You have a faded family portrait. A cracked wedding photo. A scan full of scratches, stains, and missing chunks.
You want it to look like it was taken yesterday.
One damaged scan in. One clean, restored photograph out.
Run now on Floyo!
Why Qwen Image Edit 2511 Is Different
🎨 Prompt-Driven Restoration
Most restoration tools apply fixed filters. Qwen Image Edit 2511 understands what you want in plain language.
You describe the damage. The model reconstructs the image. It doesn't just sharpen or denoise it reasons about what the original scene probably looked like and rebuilds it. The result:
scratches, tears, and stains removed
missing areas filled in with context-aware reconstruction
faces and key details kept recognizable
🧠 What the 2511 Update Changes
The 2511 version specifically improves identity and detail stability across edits.
That matters a lot for photo restoration. Earlier versions sometimes drifted faces changed slightly, backgrounds shifted. The 2511 update keeps the people in your photo looking like themselves after the repair. This means:
face identity stays consistent through heavy restoration
fine details survive the edit
original composition is respected
📸 What It Can Actually Fix
The model handles real-world damage well:
dust, grain, scratches, creases, tape marks, and paper tears
fading and yellowing from age
low contrast and blurred subjects
small missing areas, cracks, and edge damage
optional: realistic colorization of black-and-white photos
⚡ Fast With a Lightning LoRA
Pair it with a Lightning or fast LoRA and run in 4–8 sampling steps. Quick enough to batch-restore multiple photos in one session.
Key Inputs
Your Damaged Photo
Scan or photograph the original print at the highest resolution you can. The model works from what it's given. A higher-quality input scan produces a better restoration.
Works well with:
faded or yellowed prints
photos with surface scratches, dust, or creases
torn or partially missing edges
blurry or low-contrast originals
black-and-white photos you want colorized
Works less well with:
photos where large central areas (faces, main subjects) are completely missing
extremely low-resolution scans under 300px
heavily overexposed or fully burnt-out regions
Prompt
Describe the damage and what you want preserved. Be specific.
Default restoration prompt: "restore and repair this old damaged photo, remove stains, scratches, lines, and noise, reconstruct missing areas, keep faces and composition, natural modern look"
Variations:
Heavy damage:
"heavily damaged old photo restoration, rebuild missing areas, remove tears and water stains, preserve faces and original scene"With colorization:
"restore and colorize this black and white photo, natural realistic colors, remove scratches and grain, keep original faces and composition"Minimal touch:
"lightly restore this old photo, remove dust and scratches, improve sharpness and contrast, preserve original feel"
CFG Scale and Steps
Start at CFG 3.5, 8 steps with a Lightning LoRA
If backgrounds or details change too much, raise to CFG 4.5–5.0 and 15–20 steps
More steps = more faithful to the original structure. Use them when the fast run drifts.
Who This Is Great For
👨👩👧 Families and Personal Archivists
Restore heirloom prints without manual retouching skills. Clean up decades of damage in seconds before printing new copies for the family.
📷 Photographers and Restoration Studios
Use Qwen Image Edit 2511 as your AI starting point. Hand off the heavy lifting to the model, then do final fine-tuning in Photoshop, Krita, or GIMP.
🏛 Historians, Museums, and Educators
Restore historical photographs for exhibits, books, and documentaries. The 2511 identity stability update makes it more reliable for preserving authenticity in archival work.
⚙️ ComfyUI Pipeline Builders
Drop it into a batch workflow. Load multiple scans, run restoration, pass results to an upscaler or color grading node. Near-automatic restoration at scale.
Practical Workflow
1. Scan your photo at high resolution. Load it into the workflow.
2. Run with your restoration prompt and a Lightning LoRA at 4–8 steps. Check the result.
3. If details drifted, rerun with CFG 4.5–5.0 and 15–20 steps for tighter fidelity to the original.
4. Finish in your editor if needed. Healing brush and light cloning to fix any remaining artifacts. Crop and frame as desired.
That's it. Qwen Image Edit 2511 handles the heavy repair work. You do the final 10%.
What to Watch Out For
Large missing areas over faces. The model will attempt reconstruction but may not produce the correct person if too much of the face is gone. Use it as a starting point and finish manually.
Overexposed regions. Fully blown-out areas have no recoverable detail. The model fills them in but is essentially guessing. Check results carefully.
Colorization accuracy. Automatic colorization is good but not perfect. Skin tones and natural environments tend to be accurate. Clothing colors are a guess unless you specify them in your prompt.
Scan quality is the ceiling. A low-resolution or poorly lit scan limits what the model can do. Invest time in a clean scan first.