I motified the motion workflow to make it work with GGUF and added in keeping original audio. It does a good job of keeping both original audio and motion while changing the style. You could even use this to inpaint objects into the original video.
Models:
LTX23_audio_vae_bf16.safetensors
gemma_3_12B_it_fp4_mixed.safetensors
ltx-2.3_text_projection_bf16.safetensors
ltx-2.3-22b-distilled-Q5_K_M.gguf
ltx-2.3-22b-distilled-lora-384.safetensors
LTX23_video_vae_bf16.safetensors
ltx-2.3-22b-ic-lora-union-control-ref0.5.safetensors
Restyling Notes:
Text to Style seems to often just make asian realistic styles
I added flux 9b to take the first frame and restyle it for better results.
You can upload you own starting image which will apply the motion to that image.
Description
FAQ
Comments (6)
Hi, I'm using your workflow and the generated video is completely black (just a black screen with no image or content).
Could you please tell me what might be causing this and how to fix it?
Any help would be greatly appreciated!
Not sure about LTX, but with Qwen you get this if you're using Sage Attention.
I have updated to a v3 with better explanation and inpainting using flux 9b. The best results are providing your own first image. Even better if it looks similar to the original just a different style etc. Can you provide the log message / error? I had issues wiht Sage Attention myself using Qwen Edit due to Sage. Consider also my new workflow if that helps?
It is not clear how to change styles... with Loras or with prompt? In my test comes out always asian characters...
You are right, I pushed this out fast without clear explination of how it works. The best is to run it through flux klein 9b as an exact copy of the first frame in another style. I will wish an update soon and a better explination thats for pointing this out!
I have added v3. A starting image similar to the original video works best. Klein 9b can restyle the first frame or you can uploaed your own first reference image for better results.