You generated a great image at 768px — and now you need it at 4K for a poster, a thumbnail, or a client deliverable. A plain upscaler will make it bigger and blurrier. SUPIR (Scaling-Up Image Restoration) is different: it is a diffusion-based restoration model that reimagines detail as it enlarges, hallucinating plausible skin pores, fabric weave, and foliage where the source only had a hint. Run inside ComfyUI, it turns rough drafts and low-res sources into genuinely poster-ready images.
This guide covers how SUPIR works, the ComfyUI setup, and how to keep it faithful instead of hallucinating a different picture.
What Makes SUPIR Different from ESRGAN
Classic upscalers (ESRGAN, 4x-UltraSharp, and friends) are fast and deterministic — they sharpen and interpolate, but they cannot invent detail that is not implied by the pixels. SUPIR is generative: it uses a large diffusion backbone guided by your image and a text prompt to synthesize new, believable detail during the upscale. The prompt is the key idea people miss — you are describing the content so the model restores the right details.
The tradeoff: SUPIR is heavier and slower, and if you push it too hard it will "improve" your image into something subtly different. Control is the whole game.
Installing SUPIR in ComfyUI
Install ComfyUI-SUPIR (the Kijai wrapper) via ComfyUI Manager or by cloning into custom_nodes/.
Download the SUPIR model files the node requires and place them where the repo specifies. SUPIR sits on top of an SDXL base model, so you also need an SDXL checkpoint in models/checkpoints/ (a realistic SDXL model for photos; an illustration SDXL model for art).
Restart ComfyUI and load the example workflow from the repo.
The Workflow
Load Image — your source (any resolution; SUPIR shines on small inputs).
SUPIR Model Loader + your SDXL checkpoint loader.
SUPIR conditioning — this takes a positive prompt describing the image content ("a portrait of a woman, natural skin texture, sharp eyes, detailed hair") and a negative ("blurry, plastic, oversmoothed, artifacts").
SUPIR sampler — the core restoration pass.
VAE Decode > Save Image.
The Settings That Control Fidelity
Upscale factor: 2x is safe; 4x is dramatic but needs more care. For extreme jumps, upscale in two passes.
Denoise / restoration strength: the master dial. Low stays faithful to the source; high invents more (and drifts more). Start in the middle and adjust while watching a face or text region.
CFG: moderate. Too high bakes in artifacts and oversharpening.
Prompt specificity: accurate content descriptions guide better restoration; a wrong prompt makes SUPIR add the wrong details.
Tiling: SUPIR processes in tiles to manage VRAM — leave tiling on for large outputs.
Keeping It Faithful
The most common complaint — "SUPIR changed my subject's face" — comes from too much restoration strength plus a vague prompt.
Describe what is actually there, not what you wish were there.
Lower the restoration strength until identity is stable, then nudge up for texture.
For faces, a face-detailer pass or a lower strength around the face preserves identity while still sharpening.
Compare against the source at 100% zoom on key regions before accepting the result.
Optimization: Speed and VRAM
VRAM: SUPIR is memory-hungry. Use FP8 where the node supports it, keep tiling on, and lower the tile size if you OOM. On 8–12 GB, upscale at 2x and in tiles.
Speed: it is slow by design — batch it. Generate and cull at low resolution, then SUPIR only the keepers.
When to use something lighter: if you just need clean 2x with no new detail (line art, logos, screenshots), a fast ESRGAN model or Ultimate SD Upscale is the better, cheaper tool. Reserve SUPIR for photographic detail and dramatic enlargements.
A Two-Stage Pro Pipeline
A robust finishing chain many creators use: (1) generate at native model resolution, (2) run Ultimate SD Upscale (tiled, low denoise) to 2x for clean base detail, then (3) SUPIR for the final creative detail pass. You get sharpness without letting the generative step run away with the whole image.
Conclusion
SUPIR is the difference between "bigger" and "better." Treat it as a guided restoration — accurate prompt, controlled restoration strength, tiling for VRAM, and a fidelity check at 100% — and it will turn low-res drafts into deliverables. Keep a fast ESRGAN pass for the jobs that do not need invented detail, and save SUPIR's horsepower for the hero images that do.
Downloadable Workflow & References
|
0 Comments