Fixing Texture Loss When Moving SDXL Renders to Cloud Video Models
Suspicious_Pizza9529 · reddit · 2026-10-02
A creator shares a reproducible workflow for keeping texture fidelity when moving locally generated SDXL images into cloud video generators.
Problem: Local SDXL with a multi-ControlNet stack (Depth 0.8, Canny 0.5) locks composition, pose and lighting well, but web-based generators hallucinate, misread depth and over-smooth fine detail.
Failed attempts: Adding "preserve micro-textures, heavy film grain" to the I2V prompt only produced random noise and smoothed faces; dropping motion scale gave a stiff, uncinematic pan.
Diagnosis: Two layers — the flattened PNG handoff loses spatial data, and the prompt structure forces the engine to re-imagine texture.
Working parameters:
- SDXL init pass: 832x1216, CFG 4.5, 30 steps, DPM++ 2M Karras, Depth 0.8 + Canny 0.5
- MiniMax H3 render prompt: cinematic lighting, volumetric fog, rim light, raytraced reflections, static geometry, preserving micro-textures
The author switched to MiniMax Design locally plus the MiniMax H3 cloud model, using an "Environment Lock" that feeds in environment lighting data.
More from Multimodal
- Blogger turns his writings into a punk song with Claude lyrics and Suno music — technollama · 2026-10-02
- A punk song about open weights, made with Claude lyrics and Suno, casting llamas as the villains — technollama · 2026-10-02
- Fable 5.5 reportedly launching next week — iruletheworldmo · 2026-10-02
- One AI clip fakes an entire 40-person VFX shoot in 30 seconds — anthara_ai · 2026-10-02
- Open-source TostAI turns raw video into 8-direction game-ready sprite sheets, fully local — camenduru · 2026-10-02
- Qwen Image 2.1 polaroid experiments: three full prompts for realistic vintage snapshots — CommonEye1834 · 2026-10-02