[ H3 ] ref2va looks dev
I'm really interested in pushing H3 in that direction, but ref2va gen always looks bad - dark, high contrast, and oversaturated.
Proof1 community thread
Key points
- SeeDance videos look much more realistic and softer.
- I think with the H3 FL2VA workflow you can do it much better as long as you have a good start image where you control the look.
- Of course, please share your idea on this, how you get better result with ref2va.
- 2 nodes I am using "Image Detail Contrast" and "Orion4D Tone Mapping".
Sources (1)
- [1][ H3 ] ref2va looks devr/StableDiffusion (top, day) · Oct 22, 04:30 PM
I'm really interested in pushing H3 in that direction, but ref2va gen always looks bad - dark, high contrast, and oversaturated.
SeeDance videos look much more realistic and softer.
Extractive summary: sentences quoted from the sources.
Before this
- Oct 22, 2026Running Frontier-Class AI Models Across 8 NVIDIA DGX Spark Nodes
- Oct 21, 2026Claude’s personal agent
- Oct 21, 2026LoRA Training for Fibo 2.0 Added
- Oct 21, 2026Original character LoRA for MiniMax H3 learnings. Workflows included.
- Oct 20, 2026How I built my ‘everything’ app
- Oct 20, 2026[AINews] DeepSeek DeepInfer: Open Source SOTA Inference running 155T toks/day/cluster making $22m/day at ~80% margin for a 81 day payback
