Minimax H3 New VideoToVideo Animation Pipeline Using SolarRenderer + FastH3-Omni + Singularity V2 (Patreon)
Downloads
Content
Video : https://youtu.be/1tETo9M6OrQ
Hey everyone. New video is up, and this one is for anyone who's ever wanted real frame-by-frame control over an AI video - your camera, your character's pose, where every object lands.
The tool is SolarRenderer, a pair of LoRAs for MiniMax H3. You feed it a structure video - grey box blocks, a normal map, bounding boxes, even a stickman - and it renders that exact motion and camera into any world you want. In the video I turn a blocky rooftop jump into a modern-city Assassin's Creed style leap, with a female assassin diving off a tower onto an airbag in the plaza below. Same motion as the 3D blocks, every frame.
The fun part is how far the inputs stretch. A child's crayon drawing becomes a cartoon. A stickman becomes Robin Hood. Bounding boxes become a street with buses and pedestrians. And with the four-step LoRA, all of it samples in four steps - I paired it with FastH3-Omni and the new Singularity V2 for the action scenes and it runs fast.
One thing I show in the video that surprised me: the first frame alone can't carry the whole clip. When the 3D camera turns to an angle the first frame doesn't show, the render falls back to raw grey blocks - unless your text prompt describes that new scene in detail. Once I wrote the post-turn description properly, the grey houses turned into a modern city street with people walking. The prompt is half the pipeline.
If you've been stuck at DWpose and depth maps and want the full 3D-animation-level control, try SolarRenderer.
For Patreon Supporter, Additional content and resources for your membership post:
https://www.patreon.com/aifuturetech/posts/solarrenderer-h3-173451022
- Benji
Resources:
- SolarRenderer project: https://enderfga.cn/SolarRenderer/
- Model weights: https://huggingface.co/Enderfga/SolarRenderer
- SolarRenderer caption skill: https://github.com/Enderfga/SolarRenderer/blob/main/skills/solarrenderer-caption/SKILL.md
- comfyui-SelfLift: https://github.com/facok/comfyui-SelfLift
(paper: https://arxiv.org/abs/2609.02036)
- FastH3-Omni-Comfy: https://huggingface.co/FastVideo/FastH3-Omni-Comfy
- MiniMax H3 Singularity V2: https://huggingface.co/WarmBloodAban/Minimax-h3_Singularity
Calliope - Local-first Agentic Idea-to-video Studio https://github.com/benjiyaya/Calliope