4-step 768p V1.0 test results
We are continuously improving the visual quality of generated videos, with a particular focus on fine-detail rendering, resolution, and video duration. Below are examples generated with the 4-step v1.0 768p model. This version supports resolutions up to 768 × 1344. You can access and use the model through the LightX2V or MiniMax-H3-Turbo repository.
hi, wanted to ask what the ETA is to the ref2va ? :)thank you so much for these loras!
wonderful! im waiting for ref2va
Can't wait for ref2va to come out too!!
Thank you for the great work. It’s working very well. When I use Ref2VA, I use it with FL2VA + Ref LoRA by Kijai, and this LoRA seems to be working without any issues.
Thank you for the great work. It’s working very well. When I use Ref2VA, I use it with FL2VA + Ref LoRA by Kijai, and this LoRA seems to be working without any issues.
pruned or unpruned?
From all my personal testing 0.1 is better quality than 1.0
If you compare both loras results, you will see that 768p videos are in higher rez. So... is this counter-intuitive and are 768p made for higher rez and quality and in just 4 steps?
From all my personal testing 0.1 is better quality than 1.0
0.1 works amazing, up to 1.5mega pixel,
One more tip: please set the video shift to 6. In our tests, this produces better visual quality.
Thank you for the great work. It’s working very well. When I use Ref2VA, I use it with FL2VA + Ref LoRA by Kijai, and this LoRA seems to be working without any issues.
Yeah, this turbo LoRA works with Ref2VA with Ref LoRA by Kijai. So far, I don't see any issues.
One important note: prompts should be properly structured or expanded for better results. You can follow the official H3 prompt-writing guide, use MiniMax’s official IR model or another LLM, or try our Prompt Rewriter LoRA:
Official guide:
https://huggingface.co/MiniMaxAI/MiniMax-H3/blob/main/docs/VIDEO_PROMPT_WRITING_GUIDE_base_en.md
Prompt Rewriter LoRA:
https://huggingface.co/lightx2v/MiniMax-H3-Prompt-Rewriter-LoRA
Could you share the Prompts?
One more tip: please set the video shift to 6. In our tests, this produces better visual quality.
video shift=6,. only 4steps？ or 8steps also?
would be cool if you have workflow embedded in the sample videos 😄