Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

ComfyUI Tutorial for Beginners: LoRAs, Style Transfer & ControlNets Module 3 is where you stop pulling the lever and start directing. Prompt iteration gets you close. These are the tools that get you exactly where you wanted to land. Style, structure, and pose, set before the sampler ever runs.... What you'll learn: → How a LoRA attaches to a base model to change its output without retraining it → Why a LoRA has to match its base model family, and where to find them on Comfy Hub and Civitai → Model strength vs. CLIP strength — two dials, two different jobs → Stacking multiple LoRAs at different weights until the style is yours → Training your own: dataset, trigger word captions, step count, TrainLoRA locally → Reading the plot loss graph instead of guessing whether it worked → Appending to a LoRA you've already trained → Style transfer two ways: denoise-driven img2img vs. IP-Adapter reading style straight off a reference → ControlNets with DWPose, Depth Anything V2, and canny — composition locked before the prompt is read → Side by side: what each control map actually preserves → Fixing weak LoRAs, overcooked stacks, mismatched preprocessors, and noisy transfers Click the links below to run the workflow ⬇️show more

ComfyUI

57,858 subscribers

12,761 views • 16 days ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

7 Comments

Qtum.ai16 days ago

48 minute video. I wonder if I can finish this before another major AI release happens.

ComfyUI16 days ago

Stay tuned!

Olivier Heckendorn16 days ago

Where are all the tutorials ??

ComfyUI16 days ago

Our YT:

ComfyUI16 days ago

Run the workflow:

bee16 days ago

Doug can you help me with this one. I put the melon wedge in nano banana and asked it for a rectangular chisel. Got a Result in seconds. Can i do the same comfy and zimage turbo ? i need a way to control where the chisel will be and in what shape. So far nothing comes close

Marc16 days ago

Good breakdown; for video gen in the same setup, running Wan 2.2 at Q4 with 4 steps and no LoRA already gets clean 576x976 output before you even touch base model weights, worth testing that baseline first.

Related Videos

ComfyUI Tutorial for Beginners: Text-to-Image & Image-to-Image Workflows One graph, two workflows and Module 2 is where you build it node by node instead of loading someone else's. We start from a blank canvas and wire every connection, then change a single value and turn the same graph into image-to-image. What you'll learn: → The 5 core nodes, connected one at a time — checkpoint, CLIP encode, KSampler, empty latent, VAE decode → Why the data flows the way it does, not just where the wires go → Every KSampler setting: seed, steps, CFG, sampler, scheduler → Denoise from 0.9 to 0.0 — the one value that decides how much of your source survives → Locking seeds and generating takes instead of chasing one perfect image → Preview vs. save, and why every image carries the graph that made it → Reopen any result in your asset library as a live workflow → The 6 errors that break Comfy most often → How to read a red node like a sentence Click the links below to run the workflow ⬇️

ComfyUI Tutorial for Beginners: Text-to-Image & Image-to-Image Workflows One graph, two workflows and Module 2 is where you build it node by node instead of loading someone else's. We start from a blank canvas and wire every connection, then change a single value and turn the same graph into image-to-image. What you'll learn: → The 5 core nodes, connected one at a time — checkpoint, CLIP encode, KSampler, empty latent, VAE decode → Why the data flows the way it does, not just where the wires go → Every KSampler setting: seed, steps, CFG, sampler, scheduler → Denoise from 0.9 to 0.0 — the one value that decides how much of your source survives → Locking seeds and generating takes instead of chasing one perfect image → Preview vs. save, and why every image carries the graph that made it → Reopen any result in your asset library as a live workflow → The 6 errors that break Comfy most often → How to read a red node like a sentence Click the links below to run the workflow ⬇️

ComfyUI

58,853 views • 25 days ago

ComfyUI Tutorial for Beginners: Full Node Graph Basics It's a production-grade generation backend with an adaptable front end and Module 1 is where the node graph stops being intimidating. We boot up Comfy Cloud, tour the full UI, and build your first workflow from the ground up. What you'll learn: → Comfy Cloud vs. Desktop - which to use and when → Navigating the graph: pan, zoom, frame, select → The 5 core nodes behind every text-to-image workflow → Checkpoints, LoRAs, VAEs, CLIP - matching model versions → Get/Set nodes + reroutes to kill the spaghetti → Subgraphs: complex workflows collapsed into one node → App Mode: a front end for non-technical collaborators → Templates vs. Blueprints vs. Workflows → The 7 node categories that make the whole library click No installs. No missing models. Open it and generate.

ComfyUI Tutorial for Beginners: Full Node Graph Basics It's a production-grade generation backend with an adaptable front end and Module 1 is where the node graph stops being intimidating. We boot up Comfy Cloud, tour the full UI, and build your first workflow from the ground up. What you'll learn: → Comfy Cloud vs. Desktop - which to use and when → Navigating the graph: pan, zoom, frame, select → The 5 core nodes behind every text-to-image workflow → Checkpoints, LoRAs, VAEs, CLIP - matching model versions → Get/Set nodes + reroutes to kill the spaghetti → Subgraphs: complex workflows collapsed into one node → App Mode: a front end for non-technical collaborators → Templates vs. Blueprints vs. Workflows → The 7 node categories that make the whole library click No installs. No missing models. Open it and generate.

ComfyUI

14,876 views • 1 month ago

ComfyUI for Animation: Concept Art to Any Angle Concept art → 3D proxy → any angle you want. One graph, two methods. Step 1: load your concept art. Tripo generates a 3D model and saves it as a GLB. Drag it into Load 3D and set material mode to Normal. Step 2 (stills): rotate the viewport to your angle. Load 3D outputs the view, and GPT Image 2 renders it back as a polished on-style illustration. Step 3 (camera moves): record a path in the viewport. That clip becomes the motion reference for Seedance 2.0, which generates the move through the stylized scene. Click the link below to run the workflow ⬇️

ComfyUI for Animation: Concept Art to Any Angle Concept art → 3D proxy → any angle you want. One graph, two methods. Step 1: load your concept art. Tripo generates a 3D model and saves it as a GLB. Drag it into Load 3D and set material mode to Normal. Step 2 (stills): rotate the viewport to your angle. Load 3D outputs the view, and GPT Image 2 renders it back as a polished on-style illustration. Step 3 (camera moves): record a path in the viewport. That clip becomes the motion reference for Seedance 2.0, which generates the move through the stylized scene. Click the link below to run the workflow ⬇️

ComfyUI

40,683 views • 8 days ago

MiniMax H3 is native in ComfyUI, day zero with the weights. Model highlights: → Text-to-video, prompt only → Image-to-video → First-and-last-frame - control the opening frame, the closing frame, or both → Reference-to-video - carry a subject, a motion, or a voice through the clip → In-place editing - modify a shot you already have instead of regenerating it The capability MiniMax (official) leads with is multimodal context understanding, and it's what collapses those five tasks into one model. H3 takes images, audio, and video together and resolves them against a prompt that explains how they relate. Download the workflows & learn more in our blog 👇

MiniMax H3 is native in ComfyUI, day zero with the weights. Model highlights: → Text-to-video, prompt only → Image-to-video → First-and-last-frame - control the opening frame, the closing frame, or both → Reference-to-video - carry a subject, a motion, or a voice through the clip → In-place editing - modify a shot you already have instead of regenerating it The capability MiniMax (official) leads with is multimodal context understanding, and it's what collapses those five tasks into one model. H3 takes images, audio, and video together and resolves them against a prompt that explains how they relate. Download the workflows & learn more in our blog 👇

ComfyUI

234,529 views • 1 month ago

Apple built a large foundation model and fine-tuned it on multiple tasks. But they are doing something very clever: They load a single model in memory and use different adapters to specialize the model on the fly. I recorded a video to show you how to write the code to do the same thing Apple is doing. I explain everything step by step. Here is what I'll show you in the video: 1. We'll load two datasets 2. Then load a large model 3. Then, we'll fine-tune the model on both datasets I'll use LoRA to fine-tune the model. This process creates two small adapters, each specializing in solving one of the datasets. The base model's original parameters will remain unchanged. From here: 4. We'll generate a list of tasks 5. We'll load the correct adapter to solve each task The large model I'm using needs 346 MB of memory, but I only need to load it once. Each adapter is only 2.7 MB. I only need to load the base model once and pair it with any of the fine-tuned adapters. Minimum memory footprint and I can solve multiple tasks. Hope this helps!

Apple built a large foundation model and fine-tuned it on multiple tasks. But they are doing something very clever: They load a single model in memory and use different adapters to specialize the model on the fly. I recorded a video to show you how to write the code to do the same thing Apple is doing. I explain everything step by step. Here is what I'll show you in the video: 1. We'll load two datasets 2. Then load a large model 3. Then, we'll fine-tune the model on both datasets I'll use LoRA to fine-tune the model. This process creates two small adapters, each specializing in solving one of the datasets. The base model's original parameters will remain unchanged. From here: 4. We'll generate a list of tasks 5. We'll load the correct adapter to solve each task The large model I'm using needs 346 MB of memory, but I only need to load it once. Each adapter is only 2.7 MB. I only need to load the base model once and pair it with any of the fine-tuned adapters. Minimum memory footprint and I can solve multiple tasks. Hope this helps!

Santiago

84,747 views • 2 years ago

$QVAC SDK will support in 0.9.0 (gonna be release in ~10 days) LoRA fine-tuning directly on-device, letting developers customize LLMs with their own data without sending anything to the cloud. You just load a base model, point it at your training dataset, and get a lightweight LoRA adapter back — all running locally. The fine-tuned model can then be used for inference immediately, with no extra setup. Why it matters: LoRA (Low-Rank Adaptation) fine-tuning lets you specialize a general-purpose language model for your specific use case — like matching a brand's tone, mastering domain terminology, or following a particular output format — using a fraction of the compute a full fine-tune would require. QVAC handles the entire workflow locally: dataset preparation, training with configurable hyperparameters, checkpoint saving, and seamless inference with the resulting adapter. Your data never leaves the device. The developer experience: Fine-tuning with QVAC is as simple as calling "sdk.finetune()" with your dataset and a few hyperparameters. Training runs entirely on your local hardware, produces a compact LoRA adapter file, and supports pause/resume so you can stop a job and pick it back up without losing progress. The result plugs straight into QVAC's inference pipeline — no model conversion, no deployment step, just immediate local completions with your fine-tuned model.$

QVAC SDK will support in 0.9.0 (gonna be release in ~10 days) LoRA fine-tuning directly on-device, letting developers customize LLMs with their own data without sending anything to the cloud. You just load a base model, point it at your training dataset, and get a lightweight LoRA adapter back — all running locally. The fine-tuned model can then be used for inference immediately, with no extra setup. Why it matters: LoRA (Low-Rank Adaptation) fine-tuning lets you specialize a general-purpose language model for your specific use case — like matching a brand's tone, mastering domain terminology, or following a particular output format — using a fraction of the compute a full fine-tune would require. QVAC handles the entire workflow locally: dataset preparation, training with configurable hyperparameters, checkpoint saving, and seamless inference with the resulting adapter. Your data never leaves the device. The developer experience: Fine-tuning with QVAC is as simple as calling "sdk.finetune()" with your dataset and a few hyperparameters. Training runs entirely on your local hardware, produces a compact LoRA adapter file, and supports pause/resume so you can stop a job and pick it back up without losing progress. The result plugs straight into QVAC's inference pipeline — no model conversion, no deployment step, just immediate local completions with your fine-tuned model.

Paolo Ardoino 🤖

42,413 views • 5 months ago

The open source community has been delivering on LTX 2.3 LoRAs. Fine tuning LTX 2.3 unlocks control that you can't get from closed models. Here are 7 LoRAs that can save your footage. There are too many incredible LoRAs to cover, so comment below your favorite that needs a showcase! Links to all the workflows below 👇

The open source community has been delivering on LTX 2.3 LoRAs. Fine tuning LTX 2.3 unlocks control that you can't get from closed models. Here are 7 LoRAs that can save your footage. There are too many incredible LoRAs to cover, so comment below your favorite that needs a showcase! Links to all the workflows below 👇

ComfyUI

232,229 views • 4 months ago

Billions of $$ raised to generate images, and none of them let you actually edit what you get. You type a prompt, get something close, try to fix one thing, and the whole image regenerates. Now 50% of what was working is gone. So you prompt again. And again. Stuck in prompt doom loops, burning tokens every single time. The output is always a flat PNG, limiting what you can do with it. Lica World actually reads your image and breaks it into structured, editable layers. You go in, change what you need to change, and everything else stays exactly where it was. Route each layer to the right model or a capable human. You don't need to regenerate from scratch or pay the token tax to fix a font color. Enterprises publishing creatives across site, social, and email are already using Lica to own their model and own their output. We're grateful to be supported by Accel, Amjad Masad, Srinivas Narayanan, , Village Global, and Michele Catasta to build the editing layer that AI image gen has been missing. Want early access? Check the next thread below.

Billions of $$ raised to generate images, and none of them let you actually edit what you get. You type a prompt, get something close, try to fix one thing, and the whole image regenerates. Now 50% of what was working is gone. So you prompt again. And again. Stuck in prompt doom loops, burning tokens every single time. The output is always a flat PNG, limiting what you can do with it. Lica World actually reads your image and breaks it into structured, editable layers. You go in, change what you need to change, and everything else stays exactly where it was. Route each layer to the right model or a capable human. You don't need to regenerate from scratch or pay the token tax to fix a font color. Enterprises publishing creatives across site, social, and email are already using Lica to own their model and own their output. We're grateful to be supported by Accel, Amjad Masad, Srinivas Narayanan, , Village Global, and Michele Catasta to build the editing layer that AI image gen has been missing. Want early access? Check the next thread below.

Priyaa

1,441,613 views • 6 months ago

ComfyUI is now the first and only commercial license distributor for MiniMax (official)'s generative media models. MiniMax open weights have been downloaded through Comfy close to 20 million times. Not one of those downloads paid the lab a cent. That is the open-weight problem. Training frontier models is expensive. Adoption is not revenue. If open models are going to keep pace with the biggest closed labs, the labs need a way to get paid. So we built the bridge. Enterprises, studios, and agencies license through us. The labs get incentive to keep researching. Independent creators keep building exactly as they do now, funded by the organizations with budgets. A license only gets you the model. Our Forward Deployed Creatives embed with your team, train LoRAs on your style, build production workflows in your environment, and leave your people able to run and extend all of it. What they build is yours to keep. MiniMax is where we start, not where we stop. My assertion is that open-weight labs that survive the next two years will not be the ones with the best benchmarks. They will be the ones that solve distribution to paying customers first. Try for Free | Request a License today ⬇️

ComfyUI is now the first and only commercial license distributor for MiniMax (official)'s generative media models. MiniMax open weights have been downloaded through Comfy close to 20 million times. Not one of those downloads paid the lab a cent. That is the open-weight problem. Training frontier models is expensive. Adoption is not revenue. If open models are going to keep pace with the biggest closed labs, the labs need a way to get paid. So we built the bridge. Enterprises, studios, and agencies license through us. The labs get incentive to keep researching. Independent creators keep building exactly as they do now, funded by the organizations with budgets. A license only gets you the model. Our Forward Deployed Creatives embed with your team, train LoRAs on your style, build production workflows in your environment, and leave your people able to run and extend all of it. What they build is yours to keep. MiniMax is where we start, not where we stop. My assertion is that open-weight labs that survive the next two years will not be the ones with the best benchmarks. They will be the ones that solve distribution to paying customers first. Try for Free | Request a License today ⬇️

Yoland Yan

14,903 views • 28 days ago

72 hours in and GPT-5.6 Sol is still running without a single follow-up prompt. The task was to build the most comprehensive map possible of the physical infrastructure powering AI, submarine cables, data centers, power plants, substations and electricity grids. The only thing given was a simple loop: → Find the biggest gap → Research it → Build it → Test it → Publish it → Repeat Most people expected this to stall after a few hours. It didn't. The Mac restarted mid-run and Codex picked up exactly where it left off without losing anything. What this actually proves is that the prompt matters less than the goal. The smarter the model, the more room you need to give it to make its own decisions. Constrain it with steps and you get a worse version of what it could do on its own.

72 hours in and GPT-5.6 Sol is still running without a single follow-up prompt. The task was to build the most comprehensive map possible of the physical infrastructure powering AI, submarine cables, data centers, power plants, substations and electricity grids. The only thing given was a simple loop: → Find the biggest gap → Research it → Build it → Test it → Publish it → Repeat Most people expected this to stall after a few hours. It didn't. The Mac restarted mid-run and Codex picked up exactly where it left off without losing anything. What this actually proves is that the prompt matters less than the goal. The smarter the model, the more room you need to give it to make its own decisions. Constrain it with steps and you get a worse version of what it could do on its own.

0xMarioNawfal

61,751 views • 2 months ago

the best advice a Susquehanna quant ever got came from his manager in his first week. it's the entire defense against overfitting, in two sentences. six years on SIG's equity options desk, hundreds of quant interviews conducted: you've got two variables. you want the relationship between them. you're about to plot them. "before you actually observe the pattern, you want to build an idea of what you expect to see." then one of two things happens: "it's either going to match reality, and then you know that your toy model of how the world works is correct. or it's not." "and in cases where it's not, you need to take a step back and figure out why and what's missing." that's the whole thing. write down what you expect before you look. the hypothesis has to exist before the chart does.

the best advice a Susquehanna quant ever got came from his manager in his first week. it's the entire defense against overfitting, in two sentences. six years on SIG's equity options desk, hundreds of quant interviews conducted: you've got two variables. you want the relationship between them. you're about to plot them. "before you actually observe the pattern, you want to build an idea of what you expect to see." then one of two things happens: "it's either going to match reality, and then you know that your toy model of how the world works is correct. or it's not." "and in cases where it's not, you need to take a step back and figure out why and what's missing." that's the whole thing. write down what you expect before you look. the hypothesis has to exist before the chart does.

Goshawk Trades

17,825 views • 2 months ago

Elon Musk, absolute leader of the AI race with Grok Bot, and it's not a joke anymore. Ultimate guide on god-mode setup of Grok Bot, the org chart that runs while you sleep, step by step: A Chief of Staff sits in the middle with no tools of its own, BUT it reads the outcome you gave it, picks who does what, and never does the work itself. That one rule is why it never turns into the bottleneck you hired it to remove. → Researcher pulls real sources and tracks what's actually moving, not what sounds true → Writer turns that into finished copy while the research is still in the room → Visualiser gets three reference visuals once, then ships everything in that style forever → Analyst reads what performed and tells the rest of the team what to stop doing → Scheduler owns timing and holds the queue → Publisher actually ships What makes it different from every AI tool you've used: each bot gets its own computer in the cloud, its own browser, its own files, and they all share one memory. So the research is already sitting inside the draft before the draft starts. Nothing gets copy-pasted between tabs, nothing waits on you to approve step four of nine. And you never write a workflow for it. You hit record, do the job once the way you actually do it, stop. It pulls out the steps, saves them as a skill, and puts it on a schedule. The shape you're aiming for on every bot: everything reversible finished, nothing sent. 36 drafts queued, 0 published. It does all the work and stops dead at the one line only you can cross. You stop prompting. You start assigning. Full charter blocks, the approval line and the routines are in the article below ↓

Elon Musk, absolute leader of the AI race with Grok Bot, and it's not a joke anymore. Ultimate guide on god-mode setup of Grok Bot, the org chart that runs while you sleep, step by step: A Chief of Staff sits in the middle with no tools of its own, BUT it reads the outcome you gave it, picks who does what, and never does the work itself. That one rule is why it never turns into the bottleneck you hired it to remove. → Researcher pulls real sources and tracks what's actually moving, not what sounds true → Writer turns that into finished copy while the research is still in the room → Visualiser gets three reference visuals once, then ships everything in that style forever → Analyst reads what performed and tells the rest of the team what to stop doing → Scheduler owns timing and holds the queue → Publisher actually ships What makes it different from every AI tool you've used: each bot gets its own computer in the cloud, its own browser, its own files, and they all share one memory. So the research is already sitting inside the draft before the draft starts. Nothing gets copy-pasted between tabs, nothing waits on you to approve step four of nine. And you never write a workflow for it. You hit record, do the job once the way you actually do it, stop. It pulls out the steps, saves them as a skill, and puts it on a schedule. The shape you're aiming for on every bot: everything reversible finished, nothing sent. 36 drafts queued, 0 published. It does all the work and stops dead at the one line only you can cross. You stop prompting. You start assigning. Full charter blocks, the approval line and the routines are in the article below ↓

Miraqle

87,098 views • 1 month ago

Agentic Mode is taking shape before your eyes. Almost 10,000 players have already secured their tickets for the road to the $5,000,000 finale. You can already see where this is going. A new layer of gameplay is coming together inside Clash of Coins. Less manual busywork. More squadbuilding and decisions that set your Heroes in motion and let them fight according to your logic. A fundamentally new way to play. You become the commander - set the direction, build the system and watch it start fighting in your style. Built on Base. And it's already shaping up to be one of the most ambitious agentic launches. Get ready. Link below.

Agentic Mode is taking shape before your eyes. Almost 10,000 players have already secured their tickets for the road to the $5,000,000 finale. You can already see where this is going. A new layer of gameplay is coming together inside Clash of Coins. Less manual busywork. More squadbuilding and decisions that set your Heroes in motion and let them fight according to your logic. A fundamentally new way to play. You become the commander - set the direction, build the system and watch it start fighting in your style. Built on Base. And it's already shaping up to be one of the most ambitious agentic launches. Get ready. Link below.

Clash of Coins

101,159 views • 5 months ago

As I promised yesterday, I'll briefly explain LoRA training and share a workflow I made so you can do it quickly. First, let me answer a very common question: 'Why train LoRAs when we have such advanced models?' Even though we have incredibly advanced models now (like NBP), we still can't always get them to do specific things we want. Simplest example: the spritesheet LoRA I made the other day. I generated 1000 images with Nano Banana and only 100 were what I wanted. The LoRA I trained using those 100 images gives me nearly 100% consistent results. Second point is cost and speed. With LoRA, we can cut costs by 4-5x. And while doing that, we're generating 4-5x faster. How many images do you need for a good LoRA? This depends on your LoRA's complexity. For example, when I training the spritesheet LoRA, even though I used 100 images, I didn't include buildings in the training data, so this LoRA doesn't work for buildings. So think about your LoRA's use cases and add examples for as many use cases as possible to improve quality. What are paired images and how to train LoRAs for image-editing? When training LoRAs for image editing on fal, we call each edit example paired images - one with _start suffix, one with _end suffix. For example, if you're training a background remove LoRA, the unedited original photo will be your '_start' image. The image with background removed will be the '_end' image. Simply put: images we want to edit or use as reference get _start, target images we want to achieve get '_end'. Important: save both images with the same name. Like image332_start.jpg and image332_end.jpg. This way the system knows which images pair together. What about training LoRAs for models with multiple image inputs? Same logic. We still use _start and _end suffixes, but with one difference. Since there are multiple input images, we can number them: _start, _start1, _start2. Example: start images, 1st image = Woman portrait (image35_start.jpg) 2nd image = Glasses photo (image35_start1.jpg) 3rd image = Hat photo (image35_start2.jpg) Output image = portrait of woman wearing glasses and hat (image35_end.jpg) Can we do more detailed captioning? Yes. Similarly, you can improve training quality by creating a txt file for each set with the caption inside. Example: create image35.txt and write: 'Recreate the image by putting the glasses from the second image and the hat from the third image on the woman in the first image.' What are Steps? How many should I use? What's Learning Rate? Steps determines how many times the model sees and processes your training data (your images). Each step, the model learns a bit more. But as steps increase, so does the risk of overfitting. So there's no real default. But for a simpler LoRA with 20 paired images, 1000 steps is ideal. Here's a metaphor for the Steps and Learning Rate relationship: Imagine you have a balloon. Our goal is to inflate it to the optimal size. Steps = How many times we blow into the balloon Learning rate = How hard we blow each time If we blow too softly, we need to blow many more times. If we blow too hard, we risk popping it quickly and can't reach optimal size. Of course training won't explode, but it won't work as intended because it wasn't trained optimally. Training's done, now what? Once training's complete, you'll have a safetensors file. Every model you train on fal has a LoRA inference endpoint. In that inference, add your safetensors file link to the LoRA url input, and you can use your LoRA. Thanks for the read! The workflow in the video: If I forgot anything, let me know in the replies.

As I promised yesterday, I'll briefly explain LoRA training and share a workflow I made so you can do it quickly. First, let me answer a very common question: 'Why train LoRAs when we have such advanced models?' Even though we have incredibly advanced models now (like NBP), we still can't always get them to do specific things we want. Simplest example: the spritesheet LoRA I made the other day. I generated 1000 images with Nano Banana and only 100 were what I wanted. The LoRA I trained using those 100 images gives me nearly 100% consistent results. Second point is cost and speed. With LoRA, we can cut costs by 4-5x. And while doing that, we're generating 4-5x faster. How many images do you need for a good LoRA? This depends on your LoRA's complexity. For example, when I training the spritesheet LoRA, even though I used 100 images, I didn't include buildings in the training data, so this LoRA doesn't work for buildings. So think about your LoRA's use cases and add examples for as many use cases as possible to improve quality. What are paired images and how to train LoRAs for image-editing? When training LoRAs for image editing on fal, we call each edit example paired images - one with _start suffix, one with _end suffix. For example, if you're training a background remove LoRA, the unedited original photo will be your '_start' image. The image with background removed will be the '_end' image. Simply put: images we want to edit or use as reference get _start, target images we want to achieve get '_end'. Important: save both images with the same name. Like image332_start.jpg and image332_end.jpg. This way the system knows which images pair together. What about training LoRAs for models with multiple image inputs? Same logic. We still use _start and _end suffixes, but with one difference. Since there are multiple input images, we can number them: _start, _start1, _start2. Example: start images, 1st image = Woman portrait (image35_start.jpg) 2nd image = Glasses photo (image35_start1.jpg) 3rd image = Hat photo (image35_start2.jpg) Output image = portrait of woman wearing glasses and hat (image35_end.jpg) Can we do more detailed captioning? Yes. Similarly, you can improve training quality by creating a txt file for each set with the caption inside. Example: create image35.txt and write: 'Recreate the image by putting the glasses from the second image and the hat from the third image on the woman in the first image.' What are Steps? How many should I use? What's Learning Rate? Steps determines how many times the model sees and processes your training data (your images). Each step, the model learns a bit more. But as steps increase, so does the risk of overfitting. So there's no real default. But for a simpler LoRA with 20 paired images, 1000 steps is ideal. Here's a metaphor for the Steps and Learning Rate relationship: Imagine you have a balloon. Our goal is to inflate it to the optimal size. Steps = How many times we blow into the balloon Learning rate = How hard we blow each time If we blow too softly, we need to blow many more times. If we blow too hard, we risk popping it quickly and can't reach optimal size. Of course training won't explode, but it won't work as intended because it wasn't trained optimally. Training's done, now what? Once training's complete, you'll have a safetensors file. Every model you train on fal has a LoRA inference endpoint. In that inference, add your safetensors file link to the LoRA url input, and you can use your LoRA. Thanks for the read! The workflow in the video: If I forgot anything, let me know in the replies.

ilker

15,192 views • 8 months ago

You can spin up an AI workforce with one prompt. Here's why it works: it pulls the context the AI needs to lay the foundation. Give it: - your role - team size - the goals you want to hit. Then say "interview me." After that, build it out in four steps: 1) Start with 1 agent. - What does it feel like to work with one? 2) Make that agent proactive. - Now it's doing things on your behalf before you ask. 3) Expand to 2 agents. - One directs the other, or routes to it. - This is where you learn how context gets passed. 4) Scale the workforce. - A mission control where you watch it all move around. - Now you see what has to run in parallel, which agent never needed that tool, which one runs fine on a smaller model. - Not everything needs Opus. Sub-agents are haiku and sonnet. The catch: the prompt gets you agents with markdown files doing some stuff. Getting to the 90% plus level is iteration, and that part is specific to each person. Nobody can hand it to you. Start with one agent. Spin up one agent this week. The rest is discovery.

You can spin up an AI workforce with one prompt. Here's why it works: it pulls the context the AI needs to lay the foundation. Give it: - your role - team size - the goals you want to hit. Then say "interview me." After that, build it out in four steps: 1) Start with 1 agent. - What does it feel like to work with one? 2) Make that agent proactive. - Now it's doing things on your behalf before you ask. 3) Expand to 2 agents. - One directs the other, or routes to it. - This is where you learn how context gets passed. 4) Scale the workforce. - A mission control where you watch it all move around. - Now you see what has to run in parallel, which agent never needed that tool, which one runs fine on a smaller model. - Not everything needs Opus. Sub-agents are haiku and sonnet. The catch: the prompt gets you agents with markdown files doing some stuff. Getting to the 90% plus level is iteration, and that part is specific to each person. Nobody can hand it to you. Start with one agent. Spin up one agent this week. The rest is discovery.

The Startup Ideas Podcast (SIP) 🧃

18,006 views • 1 month ago

Enough of the fluff. Time for some real advice. Launch your GPT 5.6 Sol Pro. Use the following prompt (feel free to swap the second and third lines for something of your choice): Prove for me that the group of rational points on the elliptic curve (y^2=x^3+17) is finitely generated. I want to tailor the general proof to this particular case to narrow down my understanding to a particular case. Write down a full mathematical proof, step by step, in Leslie Lamport style. Unpack all the definitions and notions so that it is fully self-contained. An example here: Notice how Sol is spawning all the subagents to complete the task. What is "Leslie Lamport style"? It's a technique where you split your proof into short, one-line sentences (claims) that are easy to justify. In a situation where your proof is not a 1-to-1 translation of, e.g., Python code, the Leslie Lamport-style proof can be treated as such. It tremendously improves the possibility of verifying each claim and helps the model structure the arguments in a very clear way. Leslie Lamport, a legendary mathematician and computer scientist, was a promoter of this way of structuring proofs. Partly because of that, he created an extension of Donald Knuth’s typesetting system TeX called LaTeX. Here you can read about "Leslie Lamport-style proofs":

Enough of the fluff. Time for some real advice. Launch your GPT 5.6 Sol Pro. Use the following prompt (feel free to swap the second and third lines for something of your choice): Prove for me that the group of rational points on the elliptic curve (y^2=x^3+17) is finitely generated. I want to tailor the general proof to this particular case to narrow down my understanding to a particular case. Write down a full mathematical proof, step by step, in Leslie Lamport style. Unpack all the definitions and notions so that it is fully self-contained. An example here: Notice how Sol is spawning all the subagents to complete the task. What is "Leslie Lamport style"? It's a technique where you split your proof into short, one-line sentences (claims) that are easy to justify. In a situation where your proof is not a 1-to-1 translation of, e.g., Python code, the Leslie Lamport-style proof can be treated as such. It tremendously improves the possibility of verifying each claim and helps the model structure the arguments in a very clear way. Leslie Lamport, a legendary mathematician and computer scientist, was a promoter of this way of structuring proofs. Partly because of that, he created an extension of Donald Knuth’s typesetting system TeX called LaTeX. Here you can read about "Leslie Lamport-style proofs":

Bartosz Naskręcki

31,115 views • 2 months ago

Livestream Alert: Open World Models in Production: LTX.io and ComfyUI Founders Live A special live conversation with Yoland Yan ( CEO, ComfyUI ) & comfyanonymous ( Founder, ComfyUI ) and Zeev Farbman ( CEO, LTX ) Join us as we sit down with the teams behind LTX and ComfyUI to talk about building an open world model that integrates natively into the production tools creators use. We'll explore what it took to build LTXa foundational model that understands space, motion, and how the physical world operates—with synchronized video and audio generation, truly open weights, and the flexibility to fine-tune on your own data. We'll explore what it took to ship a frontier video model in the open, how LTX and ComfyUI are building the pipeline for AI-driven production together, and what it means for creators, developers, and the wider open-source ecosystem. What we'll cover: → LTX: the open world model for synchronized video + audio, and beyond → Native integration into ComfyUI workflows and production pipelines → Open world models vs. closed platforms and where things are headed → Fine-tuning LTX on your own data with LTX Trainer → Community Q&A — bring your questions

Livestream Alert: Open World Models in Production: LTX.io and ComfyUI Founders Live A special live conversation with Yoland Yan ( CEO, ComfyUI ) & comfyanonymous ( Founder, ComfyUI ) and Zeev Farbman ( CEO, LTX ) Join us as we sit down with the teams behind LTX and ComfyUI to talk about building an open world model that integrates natively into the production tools creators use. We'll explore what it took to build LTXa foundational model that understands space, motion, and how the physical world operates—with synchronized video and audio generation, truly open weights, and the flexibility to fine-tune on your own data. We'll explore what it took to ship a frontier video model in the open, how LTX and ComfyUI are building the pipeline for AI-driven production together, and what it means for creators, developers, and the wider open-source ecosystem. What we'll cover: → LTX: the open world model for synchronized video + audio, and beyond → Native integration into ComfyUI workflows and production pipelines → Open world models vs. closed platforms and where things are headed → Fine-tuning LTX on your own data with LTX Trainer → Community Q&A — bring your questions

ComfyUI

13,174 views • 2 months ago

context engineering vs graph engineering. every few months the list gets a new word and everyone treats it as a replacement for the last one. these two are not on the same list. one decides what the model sees this turn, the other decides what exists at all. the cleanest way to tell them apart is to ask what a single unit of work looks like. > context engineering is the window the window opens empty, every single time. you assemble what goes in it. the prompt, the docs, the history, the tool results. the assembling is the work. the window only grows. it never shrinks on its own, so eventually something gets dropped. usually from the middle. usually without telling you. then the turn ends and the window is thrown away. not archived, thrown away. the next turn opens empty again and you re-explain what you already explained. good context engineering is knowing what to leave out, not what to pack in. the unit of work is one window. > graph engineering is the structure the same material arrives from the same sources. instead of packing it into a window, you pull entities out of it, resolve the duplicates into one node, and write typed edges between them. nothing here is stored as text you hope to find again. it is stored as a thing with a name and its connections to other things. when the turn ends, the graph is still there. the next turn does not start from zero. it starts by querying what already exists, and the query walks edges instead of guessing at similarity. good graph engineering is deciding what counts as the same thing twice. the unit of work is one relationship. > they are not alternatives the graph is what refills the window. context engineering decides what fits. graph engineering decides what there is to choose from. remove the graph and every session starts blind. remove the context work and the best structure in the world arrives as an unreadable dump. that also tells you which one broke. the answer drifted from what you actually said, or forgot something from this same session. that is the window. the answer is coherent but invents a connection that does not exist, or cannot join two facts it has clearly seen. that is the structure. people debug the prompt because the prompt is the easiest thing to edit. it keeps taking the blame for failures that live a layer down. save this - then read the full breakdown below

context engineering vs graph engineering. every few months the list gets a new word and everyone treats it as a replacement for the last one. these two are not on the same list. one decides what the model sees this turn, the other decides what exists at all. the cleanest way to tell them apart is to ask what a single unit of work looks like. > context engineering is the window the window opens empty, every single time. you assemble what goes in it. the prompt, the docs, the history, the tool results. the assembling is the work. the window only grows. it never shrinks on its own, so eventually something gets dropped. usually from the middle. usually without telling you. then the turn ends and the window is thrown away. not archived, thrown away. the next turn opens empty again and you re-explain what you already explained. good context engineering is knowing what to leave out, not what to pack in. the unit of work is one window. > graph engineering is the structure the same material arrives from the same sources. instead of packing it into a window, you pull entities out of it, resolve the duplicates into one node, and write typed edges between them. nothing here is stored as text you hope to find again. it is stored as a thing with a name and its connections to other things. when the turn ends, the graph is still there. the next turn does not start from zero. it starts by querying what already exists, and the query walks edges instead of guessing at similarity. good graph engineering is deciding what counts as the same thing twice. the unit of work is one relationship. > they are not alternatives the graph is what refills the window. context engineering decides what fits. graph engineering decides what there is to choose from. remove the graph and every session starts blind. remove the context work and the best structure in the world arrives as an unreadable dump. that also tells you which one broke. the answer drifted from what you actually said, or forgot something from this same session. that is the window. the answer is coherent but invents a connection that does not exist, or cannot join two facts it has clearly seen. that is the structure. people debug the prompt because the prompt is the easiest thing to edit. it keeps taking the blame for failures that live a layer down. save this - then read the full breakdown below

Hanako

19,160 views • 1 month ago

Ok, so the new Midjourney Style Tuner is here and it's a MASSIVE update. There is so much to unpack, no way I can do it all in one post, but here is an overview. You start by typing in /tune followed by your prompt. MJ will generate a range of sample images (between 16-128) showing different visual styles based on your prompt. Then you choose your favorite images from the sample to generate a unique code you can use to customize the look of future jobs. (link below) So for example, I just tried it with the prompt: 💬 1990s, a photo of a woman dressed in fuzzy velour, shimmering midnight hues reflecting sadness by the lake, a contemporary nightmare on cross-processed film --ar 4:5 --c 4 I selected 10 images, which generated the code zlPhgLCHDXol64kPlwSUzu97. I can then add it to my prompt using --style {code} But it makes more sense if you try it yourself, so try playing around with my Style Tuner for that prompt and generate your own code from your selections: Some quick notes: -You can control the level of style applied by adding the --s parameter to your prompt. -You can combine multiple styles together by separating with a hyphen, i.e --style code1-code2 I will be obsessing over this all day tomorrow and sharing everything I learn along the way. This is an incredible update.

Ok, so the new Midjourney Style Tuner is here and it's a MASSIVE update. There is so much to unpack, no way I can do it all in one post, but here is an overview. You start by typing in /tune followed by your prompt. MJ will generate a range of sample images (between 16-128) showing different visual styles based on your prompt. Then you choose your favorite images from the sample to generate a unique code you can use to customize the look of future jobs. (link below) So for example, I just tried it with the prompt: 💬 1990s, a photo of a woman dressed in fuzzy velour, shimmering midnight hues reflecting sadness by the lake, a contemporary nightmare on cross-processed film --ar 4:5 --c 4 I selected 10 images, which generated the code zlPhgLCHDXol64kPlwSUzu97. I can then add it to my prompt using --style {code} But it makes more sense if you try it yourself, so try playing around with my Style Tuner for that prompt and generate your own code from your selections: Some quick notes: -You can control the level of style applied by adding the --s parameter to your prompt. -You can combine multiple styles together by separating with a hyphen, i.e --style code1-code2 I will be obsessing over this all day tomorrow and sharing everything I learn along the way. This is an incredible update.

Nick St. Pierre

377,975 views • 2 years ago