正在加载视频...

视频加载失败

GPT2 + Seedance 2.0 → Yoga Flow. Controlling pose transition. (Prompt + process in thread) PROCESS: 01. GPT2 Img: Create pose sequence in 02. GPT2 Img: Create base character 03. SD2 Omni-Ref: Animating scene 04. Codex for stitching + frame interpolation While not perfect step-to-step... Still impressive.

49,335 次观看 • 6 个月前 •via X (Twitter)

39 条评论

Rory Flynn 的头像
Rory Flynn6 个月前

01. Generating Motion Sequence - Used GPT2 image to create 2 sequences - To use as reference images for SD2 - Describing the motion and transition PROMPT: Create a 12-panel yoga instruction diagram in a clean 4-column by 3-row grid. Style: black-and-white / soft grayscale instructional poster, off-white background, thin black panel borders, bold uppercase titles, black rounded number badges, simple dotted or curved motion arrows, and short coaching captions. Use one consistent female yoga model across all panels: athletic build, long wavy hair, fitted sports bra and leggings, barefoot, realistic photo-illustration style, neutral studio background. Show a classic vinyasa flow from Mountain Pose to Downward-Facing Dog. Focus less on overly technical pose detail and more on the feeling of a smooth flowing sequence, with clear movement, breath, and transitions from one step to the next. Each panel should clearly show the motion step and how it transitions into the next pose. Use arrows to show direction of movement. Keep the captions short and simple, describing the action, breath, or flow cue in a natural way. Make all 12 panels exactly the same size, evenly aligned, highly readable, and visually consistent. No color. No extra text.

Rory Flynn 的头像
Rory Flynn6 个月前

02. Generate Base Character - Cropped one of the frames for pose - Added character description - Generated with GPT2

Rory Flynn 的头像
Rory Flynn6 个月前

03. SD2 To Bring to Life - Used Omni-ref with multiple items - Base Character image as character ref - Instructional diagram for motion guidance Ex Prompt: Scene: One continuous shot - Hand-held YouTube-style video showing a woman performing a complete yoga vinyasa flow. Character: Use IMG1 as the character we're following in the scene. Character Motion: Use IMG2 to follow the motion and instructions exactly to complete her flow. The scene starts at IMG1. Direction: The woman stands in a peaceful orange yoga studio. The hand-held camera subtly pushes in on her as she begins her flow. It captures the moment in tight composition as she moves from mountain pose to upward salute. The camera begins to pull back out as she transitions from upward salute to swan dive and stays focused on her body as she transitions from swan dive into a forward fold. The camera begins to move right to a side profile as she plants her palms and steps back into the high plank position. The camera drops lower and pushes in towards her face for a focused close-up as she moves from chaturanga lower to upward-facing dog. The camera again pulls out, showing her full body as she transitions to the downward-facing dog position. Finally, she holds for a deep inhale and exhale in that position. Overall Tone: Her movements are all natural and fluid, controlled and directed, strong with natural body movements and subtle readjustments as she gets her balance. SoundFx: peaceful ambient hum as we hear her inhale and exhale in a rhythmic fashion with each passing pose. No music.

Rory Flynn 的头像
Rory Flynn6 个月前

Overall - It didn't match every pose verbatim - If i extended this to 3 videos and shortened prompts - It probably would listen - Still extremely impressive from both GPT + SD2 - The diagram as a guide is repurposeable

Marco 的头像
Marco6 个月前

Character vitiligo is cool. She looks very realistic

Rory Flynn 的头像
Rory Flynn6 个月前

Wanted to see if it would hold the vitiligo through the sequence

Arsalan 的头像
Arsalan6 个月前

This is a really clever way to make Seedance 2.0 follow motion properly

AmirMušić 的头像
AmirMušić6 个月前

Man this might become something really serious 🔥

Rory Flynn 的头像
Rory Flynn6 个月前

It’s like these two were meant for each other

AmirMušić 的头像
AmirMušić6 个月前

hahaha. i meant the workflow, but who cares now 😁

DavidsAIMastery 的头像
DavidsAIMastery6 个月前

Two Important Tips to ADD🔥 1. add "make the yoga posing model face 45 degree forward" so they give you more dimensional angles. 2. When you are writing the seedance prompt, YOU MUST check the pose names in the prompt with the pose names in the picture. Otherwise the video might turn out weird! (chatgpt may generate 12 poses with slight different names)

Zephyra Leigh 的头像
Zephyra Leigh6 个月前

You really showed skill

Elvisbo 的头像
Elvisbo6 个月前

Wow. Great work.

ALG∑BRA 的头像
ALG∑BRA6 个月前

why is her skin multi colored??

Ilya Pivto 的头像
Ilya Pivto6 个月前

Camera movement are great. How much did it cost to generate this video?

Ur Meer 的头像
Ur Meer6 个月前

Perfect combo

UmarAi 的头像
UmarAi6 个月前

Dope concept

Marc Lindsay 的头像
Marc Lindsay6 个月前

Okay this is Increadible so this could easily use my own images to create the same thing. Amazing.

Pelaez Design AI 的头像
Pelaez Design AI6 个月前

Wow, this is dope and inspiring. Can't wait to try it out! Thank you for sharing! 😃

Adel Bucetta 的头像
Adel Bucetta6 个月前

the honest answer is that even with a clear process, it's still hard to predict what will work and what won't the output is only as good as the input and the constraints you set around it

Alpha Batcher 的头像
Alpha Batcher6 个月前

We can do yoga in gpt image and seedance 2.0 now 🤭

Maercih 的头像
Maercih6 个月前

Innovative approach unfolding nicely.

Anissa 的头像
Anissa6 个月前

Amazing creativity man!

Julia Clark 的头像
Julia Clark6 个月前

Looks good

ZeFred.AI 的头像
ZeFred.AI6 个月前

Yes really impressive 🔥

Terry Bui 的头像
Terry Bui6 个月前

流れるヨガがとても魅力的ですね。素晴らしい技術を使っているのが感じられます。

Tabi Yume 的头像
Tabi Yume6 个月前

Anything and everything seems to be possible with lethal combination of GPT2 img and SD2. Crazy time to be living and yet we havent even scratched the surface👍

Girl AI 的头像
Girl AI6 个月前

hay quá

Ciri 的头像
Ciri6 个月前

looks like some amazing work

AI Mastery Guide 的头像
AI Mastery Guide6 个月前

'While not perfect step-to-step' is doing a lot of heavy lifting here but honestly still miles ahead of where this was 6 months ago 😂

唐清乐 的头像
唐清乐6 个月前

多工具串联的工作流比单一工具的输出有趣得多。用图像生成姿态、用视频模型让它动起来、再用代码做帧插值——每一步都不完美但组合在一起能产出单工具做不到的东西。关键是人得充当编排者,知道什么时候该用哪个工具、什么时候该手动修。

Barrak 的头像
Barrak6 个月前

Combining GPT2 for pose sequencing with Seedance for animation is a clever pipeline and the result speaks for itself.

Luis TG 的头像
Luis TG6 个月前

your are awesome for share the prompts,,,thx u

Miraaj. 的头像
Miraaj.6 个月前

Here's how ideas change into real time video clip by Seedance from Gpt image

ᴍᴜʀᴘʜʏ 的头像
ᴍᴜʀᴘʜʏ6 个月前

You have done it nicely

Marcel Velica 的头像
Marcel Velica6 个月前

Nice pipeline, this is a solid example of multi-model orchestration for controlled motion generation.

DavidsAIMastery 的头像
DavidsAIMastery6 个月前

Thanks for Sharing! Got these two awesome Yoga Demonstrations🔥 For the modification, whole step by step prompts🔽

EasyClaw 的头像
EasyClaw6 个月前

You can also use Seedance in #easyclaw And other top AI models in one platform. 200 free credits daily and now you can get more.

Somake 的头像
Somake6 个月前

The pose sequence step is what makes this work.

相关视频