It looks like Wan2.2 For-Loop Scenario, but only 1 step. Then you can use output video as input and continuew it. At any step if you don't like something you can change and retake only 1 scene. Also you can use different step count for steps.
Description
Use this version to create a sync loop.
Start from any video
End with specified image (frame)
FAQ
Comments (5)
This is very nice BUT, would be nice to have 5-6 generations on this workflow that you can bypass, with different prompts. So for example:
- generate first iteration until you're happy with it, keep seed locked, un-bypass second step, it automatically gets the last generation last frame and does its thing until you get one you like, rinse and repeat. It's like your other workflow but with more control over it. And it's important to have different prompts for each one because of the control it gives you.
Yes! Actually the BEST would be if we can do "draft" in first line. Like a small size or short video - but... I've check : ))) if I need 480x640 and test on 240x320 - it's completelly diff video (steps count give the same) In some case short-preview give me a knoledge about concept (First time I place "big snake appears from the back" and is was bigger than a woman : ))) so I change the prompt. and also in realise that I need a scenario (do what you want all night) or step-by-step (if I need a full control). I see what you mean and it's a make sence but... I don't see how to do that.
BTW - do you see a scenario version?
@asdfafafa uh (cough) that's exactly what I built in my multi-WAN-KR, lol. generate the 1st clip, if you're not happy - change seed or prompt and redo it. if clip 1 is good enable the 2nd clip which picks up from the 1st automatically and continue there (generate, check, regenerate or keep). same with 3 and 4. in the end combine on the fly and optionally interpolate. In each step you have a preview of of current clip and last clip combined to check for bad transitions or jerky cuts. What is not yet in 0.9.2 is a lora stack for each clip. It's in my local version and it works just fine to build a storyboard by using different loras for each step. Caution: if you start Comfy with "--cache none" this will not work obviously.
@gman_umscht Great job! I also start pet-project based on Comfy-backend, but too long chew the gum : )))) still nothing : )