Video wird geladen...
Video konnte nicht geladen werden
WWE 2K Gameplay Sliders | Player v AI V1 Difficulty: Hard (6-12mins) Legend (8-15mins) Gameplay/Presentation: Preference Match Options: Special: Normal Finisher: Slow Upper Midcard and Main Event: Slow & Slow More info in the comments: ⬇️⬇️⬇️ #WWE2K26
90,540 Aufrufe • vor 6 Monaten •via X (Twitter)
0 Kommentare
Keine Kommentare verfügbar
Kommentare vom Original-Post werden hier angezeigt
![GTA in Brazil 🇧🇷 All this chaos just to deliver the package to the right house! Created using Seedance 2.5 on Pollo AI Prompt: Create a 30-second ultra-realistic AAA third-person open-world gameplay sequence set in a densely populated Brazilian neighborhood inspired by Rio de Janeiro. The video must look like genuine next-generation GTA-style gameplay captured directly from a playable open-world game — not a cinematic film. Use a consistent third-person gameplay camera positioned naturally behind the protagonist, realistic player-controlled movement, seamless environmental interaction, physically accurate collisions, natural NPC behavior, detailed Brazilian surroundings, and believable game-world physics. MAIN CHARACTER: Use the uploaded character image as the exact identity reference for the protagonist. Preserve her face, facial proportions, hairstyle, hair color, skin texture, body proportions and overall identity consistently throughout the entire sequence. She is a young Korean woman in her early 20s with long black hair, wearing a casual summer outfit with a pastel-lavender top, loose white cargo pants and sneakers. Her clothing, face and hairstyle must remain completely consistent from beginning to end. OPENING — 0–3 SECONDS: Begin immediately in playable third-person view. She is casually walking along a busy Brazilian residential street. The environment instantly establishes Brazil: dense hillside homes, colorful painted walls, Portuguese storefront signs, overhead electrical cables, parked compact cars, motorcycles, pedestrians, small neighborhood businesses, tropical plants and realistic street clutter. Without warning, a motorcycle delivery rider pulls up nearby and places a sealed cardboard package on a low wall while checking his phone. The protagonist looks at it. A contextual gameplay prompt appears: “TAKE PACKAGE [E]” She looks around. She takes the package. Immediately: MISSION STARTED “DON'T GET CAUGHT WITH THE PACKAGE” 3–8 SECONDS: The delivery rider looks up. He realizes the package is gone. A red alert appears: “PACKAGE STOLEN” He points toward her and starts chasing. She freezes for half a second, realizes what is happening, then runs. The camera smoothly follows her like authentic third-person gameplay. She runs downhill through a crowded Brazilian street, narrowly passing pedestrians and motorcycles. NPCs react naturally — some step aside, one briefly looks back, another continues walking while barely acknowledging the chaos. 8–15 SECONDS: She cuts through a narrow alley between houses. The rider follows on foot. She jumps over a low concrete barrier and continues running. A motorcycle passes extremely close in front of her. She instinctively stops for a fraction of a second. A notification appears: “BAD DECISION” She immediately turns down another street. Her stamina bar visibly drops. “STAMINA LOW” She reaches a small local football court where several people are playing. A ball suddenly rolls directly into her path. She kicks it aside without stopping. The players briefly react as she runs through. 15–22 SECONDS: She reaches a steep hillside street. She looks back. The delivery rider is still chasing her. She notices an open doorway beside a tiny neighborhood shop and quickly ducks inside. The rider runs past. The screen briefly displays: “THEY LOST YOU” She quietly steps back outside. The rider suddenly appears at the other end of the street. “THEY FOUND YOU” She panics and runs again. 22–27 SECONDS: She finally reaches a small Brazilian house matching the address printed on the package. She looks at the label. Then at the house number. A subtle objective updates: “DELIVER PACKAGE” She rings the doorbell. A Brazilian woman opens the door. The protagonist hands her the package. The woman checks the address and happily accepts it. Objective instantly changes: “DELIVERY COMPLETE” 27–30 SECONDS: The exhausted delivery rider finally arrives. He stops. Looks at the protagonist. Looks at the woman. Looks at the package. He slowly realizes that the protagonist accidentally delivered it to the correct address. He is completely confused. The protagonist gives him an innocent smile and casually walks away. Final HUD notification: “MISSION COMPLETE” “CHAOS +250” The rider remains standing there, speechless. GAMEPLAY & VISUAL STYLE: Ultra-photorealistic AAA open-world game graphics with realistic human characters, detailed facial animation, physically based materials, realistic skin, individual hair strands, natural cloth movement, accurate shadows, global illumination, realistic reflections, detailed road surfaces and authentic environmental imperfections. Brazil must be visually recognizable through authentic environmental details rather than exaggerated stereotypes. Use realistic Rio-inspired hillside neighborhoods, dense residential streets, Portuguese signage, Brazilian motorcycles and compact cars, local stores, football culture, tropical vegetation, concrete walls, balconies, utility poles and tangled overhead wiring. NPCs should have varied appearances, clothing, animations and reactions. No duplicated-looking crowds. All physics must be grounded: the cardboard package has believable weight, pedestrians react naturally to collisions, motorcycles have realistic acceleration and suspension, clothing responds naturally to movement, and characters maintain consistent proportions. CAMERA: Strictly authentic third-person open-world gameplay perspective. No cinematic establishing shots. No drone shots. No slow-motion hero shots. No dramatic camera transitions. No movie-style depth of field. No impossible camera movements. The camera should behave as if a player is controlling the character in a real AAA open-world game. HUD: Use a restrained GTA-inspired gameplay HUD integrated naturally into the screen: - circular minimap - health bar - stamina bar - mission objective tracker - contextual interaction prompts - alert indicators - small gameplay notifications The interface should feel like part of an actual AAA game rather than an overlay added afterward. NEGATIVE REQUIREMENTS: No anime. No cartoon. No cel shading. No mobile-game graphics. No fantasy. No exaggerated physics. No gore. No distorted faces. No changing character identity. No changing clothing. No inconsistent package size. No duplicated NPCs. No teleporting objects. No floating objects. No rubbery character movement. No cinematic movie shots. No unrealistic facial expressions. No artificial-looking crowds. The final result should feel like a completely authentic 30-second gameplay clip from a huge, photorealistic AAA open-world game set in Brazil, funny because of the situation and gameplay mechanics, not because the visuals are exaggerated.](https://image.24vids.com/tw-2099832403305079015/amplify_video_thumb/2099832237663645696/img/gSCB4v3zHFp7htZX.jpg)
![I think I can finally report some success training a quite accurate IDM capable of recovering keystrokes from Minecraft gameplay, even in quite PvP-heavy situations. At this point the model does not only know what keys are pressed to the extent reasonably discernible, it also knows how fast it is moving in 3D space at all times, even when knockback is mixing with the self-move impulse. Now, recovering keystrokes from normal external capture footage is just about impossible. E.g. W/A/S/D does exactly nothing during partial tick frames and jumping mid-air is also equally useless, so asking the model to recover key down states is inherently unreasoanble. Mouse deltas are also completely arbitrary units, as game mouse sensitivity introduces an arbitrary scale factor into the equation. The only good option is to think carefully about your model-environment contract, and only record "logical actions", not raw keystrokes. So here's a few unfortunate lessons I had to learn in roughly this order. - Choose good units. (bad: mouse deltas, good: delta radians [yes, you will need game-internal state]) - Capture from inside the main game loop and read the game fbo to get consistent frame-action pairing. Doing post-mortem pairing is hopeless. - Carefully define when you think keystrokes actually have an effect. (jump only works on ground, when flying or in water etc.) More subtle: The key may already be down, but no tick has happened yet to actually use the value. Hence: ignore Seperate gamestate into "fast and slow-moving" components. E.g. movement is likely tick based, camera rotation is very likely updated every frame in essentially every game ever. - Think about your frame-action correspondance contract (How old is the frame in relation to the inputs you capture? Will double or tripple buffering affect you?) Think about the game loop timeline, where you are sampling, how old the data you are reading is, and where the ticks are happening around you. Language models used to simply not have a model-environment contract, but even now with the model "living" in a designated harness, the contract still boils down to formatting, and tool implementation intrinsics. While also important, it is still quite a bit more obvious because the violations are in some way shape or form reflected as text you can actually see. - ffmpeg dropping frames cummulatively screws the model the further you get into the sequence because your targets are now shifted. If you can't encode the video in real-time, too bad. - Sodium has a frames in flight system different from vanilla Minecraft, which will also offset your targets from your frames. (there goes that data...) - Models are succeptible to latency. If there is too big of a delay between action and on-screen reflection, your performance degrades. At this point I realize ~100hours of gameplay is essentially no longer usable as a dataset. You can train on this data, but all you'll get is a mushy mess. However, some good news: - Making the model predict physics gamestate scalars helps the model generalize. For instantaneous events like jump, it's unreasonable to ask the model emit a short burst of jump=true at exactly the right time, however if you also predict your current y-velocity, the model has supervision signal for the "latent" from which that onground jump becomes apparent. Recovering x/z motion is also somewhat easier than unmixing it into plausible keystrokes for inertia-heavy player controller logic. - Regressing physics gamestate scalars also seems to make your dataset "bigger". While pure keystroke classification will overfit quickly, predicting exact physics gamestate scalars forces the model to generalize more and you can tolerate far more epochs before validation loss starts to stall out. This is the only reason why it was bearable to dump 100h+ of dataset hours and replace it with ~3 hours of gameplay after the 4th revision of the file format (yeah...) and somehow still have better performance. Now, you might be asking, "isn't this brittle?" and the answer is yesn't. Frame-action correspondance matters for training, but not so much during inference. So as long as you are sampling in roughly the same interval as your training data, you aren't violating any hard contract per-se. Somewhere around the frames ticks are happening, and during training you capture various tick-capture offset relations per random chance, so nothing is too obviously wrong here. HOWEVER, you will get screwed by gui scale, shaders, resource packs, "shit that recording is 1920x1040 because somebody doesn't know fullscreen exists" and other unfortunate edge cases of reality. But I suppose this is the role of dataset size. If all those "contract violations" that a youtube video has compared to the training data are addressed, I think this is a way to turn Youtube into a labeled dataset. I could never shake the feeling that VPT is a sound idea in practice, while never having been properly executed, and I think one reason why it hasn't is because that label boostrapping part is just a pain in the butt to get right. Now, what the player is doing is of course not the only label you can extract from video, but it has to be one of the targets predicted during pretraining to "align" the pretraining objective. Some notes on the video here, the colored dots on the analog visualizer are the ground truth, while the gray dot is the model prediction. Green means correct prediction, red means incorrect prediction at that frame. Model P(key) reports how wrong the prediction is from green (0.0) to red (1.0). You will also notice that during periods of rapid slow down, left and right actions become close to irrecoverable, because there is just that little motion. And some jump actions are not predicted correctly because I got the detection condition for jump events wrong... (duh) LMB/RMB for other than sustained events (like item-consume and block break) also seem to be hopelessly irrecoverable for now. Swing was supposed to do the same thing as motion y did for jump, but its too well behaved as an increasing counter. Maybe partial-tick interpolated values work better (v5 file format then... ugh..)](https://image.24vids.com/tw-2042731804780826849/amplify_video_thumb/2042721691844120576/img/yugYQfuXXQmJpdVp.jpg)
