ComfyUI user asks how to cache latents and fix weak second-pass upscaling
Extreme_Speed6654 · reddit · 2026-07-24
A Reddit user shares a ComfyUI workflow for low-resolution latent seed hunting and asks for advice on three practical issues:
- how to keep a latent cached in memory while optionally saving it to disk only when needed,
- why the second-pass refinement still looks low quality,
- and why the upscaling stage is extremely expensive on GPU but produces disappointing results.
The user says the pipeline mostly works, but suspects the problem is the workflow design rather than the model itself, and offers to share screenshots if needed.
More from Multimodal
- Reply post shows a polished AI-generated elf portrait — azed_ai · 2026-07-24
- A Seedance 2.0 prompt maps a pirate battle shot-by-shot for cinematic video generation — techhalla · 2026-07-24
- AI-Generated Creative Short Film: Solar Flight — Luzifee-666 · 2026-07-24
- Fizgig adds per-image loss tracking, auto-recaptioning with Qwen3-VL, and resume-safe LoRA training — shootthesound · 2026-07-24
- HeyGen open-sources HyperFrames, an agent that turns prompts into frame-by-frame launch videos — Vjeux · 2026-07-24
- Codex turns one mascot brief into 30 app-specific variations in one shot — jarrodwatts · 2026-07-24