👍. Excellent tip. I was reading about that last week, to do with a client enquiry for some multi language deep fake project... Interesting. I'll definitely try it out soon.
can you please explain what you mean by "running roop" I recently purchased Topaz Video Ai and am getting more in Midjourney. I'm a novice by most standards, Thanks.
@@chadiwack yes, no problems. Roop is a lightweight, free program that allows you to select an image and granted the application is able to discern a face it will replace faces within a video of your selection with that of the image. There is also an extension for it in A1111.
I'm sure that might work ok and help in some instances. You can also just up the passes for 'remove jittery motion' within the software. So it will do multiple passes as it processes the clip. By default, it's set to 2 but you can up it to 5.
Thanks for your video. I saw another tutorial that recommended turning off the interpolation option in Runway and adding it in at the Topaz end. Unfortunately you didn't open the Runway drop down settings panel where it allows you to choose upscale and add interpolation etc, but it's definitely worth investigating.
😃 Think I've since seen the same video. I've tried both approaches and actually think I get a marginally better result with Runways Interpolation done first... then run the interpolation in Topaz... but I'm not 100% sure.
I have one simple question the man you genrated facing the camera how to make this specific character person to do some other activities in same outfits..
That's one of the big challenges at the moment that various creatives and AI companies are trying to solve. The phrase 'consistent characters' is used widely in the pursuit of creating scenes with the same character. The best approach I've found so far is to reuse the image of a character you like and use that as an image reference when generating another image in Midjourney, along with a prompt describing them in the same way as the original image, plus describe what you'd like them to be doing. You can also use the prompt, 'multiple pose sheet' when first designing the character to get various pose options and angles in a single image, which you could then try and work with in Photoshop to create your image. But it's far from straightforward. Gen-2 of Runway, with an image (no text) prompt in RunwayML gives little (no) control to what action the AI decides to bring to the footage. So it's a bit of a 'suck it and see' approach. Hopefully, this will change in the future. If you add a text prompt as well as an image at the moment, the results only take some of the style and feel of the image and won't exactly match the provided image, scene or characters. You could use Gen-1 of Runway, (video to video). Film a scene of an actor (or yourself) doing the needed motion, and use an image reference (and or text prompt) to guide what it looks like. It will then match your video motion/composition and apply the look of the image/text prompt. This is a nice workflow but the quality produced currently is not fantastic, there's often some distortion, low resolution, and the movement of mouths are quite hit n miss. You could then combine it with other tools to try and improve the quality, composite in a better-animated mouth etc etc. So whilst it's not super useful yet as this improves it will open up an approach for having more control over a scene. That said, it's impressive and some decent (albeit not perfect quality) results can be achieved. Lastly, you could take a much more involved production approach and create a 3D model of a character, bring it to life using something like Wonder Studio (where you can turn a live-action filmed actor into a CGI character and match lighting and download mocap data to use in a 3D package.. i.e. Blender).
@@shailendrarathore445 . I go through the process with Wonder Studio in this video. ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-c3L7O-0gRiQ.html . Plus will be doing it again in another one soon.
Checkout Pika Labs for a currently free alternative to runway. Plus AnimateDiff is complex but awesome and free (using your own hardware) (with ComfyUI for stable diffusion generations).
