AI IMAGE EDITING
How to Edit an Image with a Reference Photo (Free AI, Prompt-Based)
“Image‑to‑image” editing is the sweet spot between a text prompt and a starting photo: you give the AI a reference photo plus a prompt describing the change, and it edits the photo toward that description. Want to change a shirt colour, restyle a room, or add a sunset? This step‑by‑step guide shows you how — free online, no sign‑up, no watermark.
What “Image‑to‑Image” / Reference Editing Is
Most “AI image generators” build a picture from words alone. Reference editing (also called “image‑to‑image”) gives the AI a starting image — your reference photo — and a prompt that says what to change. The model then edits toward the description while keeping the parts of the photo you didn’t mention. “A person in a red shirt” stays the same person; the shirt becomes red.
That makes it powerful for things pure text‑to‑image can’t do predictably: colour swaps, style changes, background replacements, and pose‑preserving tweaks.
The Qwen‑Image‑Edit‑2511 Model (What Makes It Special)
The online generator’s Qwen‑Image‑Edit‑2511 model is built exactly for this job: it accepts a prompt and up to 2 reference images, then edits the composition described in the prompt while anchoring the unchanged areas to the references.
| Model | Input needed | Cost | Time | Reference images |
|---|---|---|---|---|
| Z‑Image‑Turbo | Text prompt only | €0.02 | seconds | 0 |
| Qwen‑Image‑2512 | Text prompt only | €0.10 | ~1 min | 0 |
| Qwen‑Image‑Edit‑2511 | Text prompt + reference image(s) | €0.12 | ~2–3 min | up to 2 |
The key difference: Qwen‑Image‑Edit is the only model here that takes reference images, so it’s the one to pick whenever you want to change an existing photo rather than paint a blank canvas.
Reference Photo Rules (Size, Count, Format)
- Max 2 reference images per request — the model is designed around one main image (optionally a second for style guidance).
- Accepted formats: JPG, PNG, WebP — the same formats the generator takes for text‑to‑image.
- Size guidance: the site automatically scales any reference larger than ~1.5 MP down to keep generation fast. If your reference is smaller than 512 ×512, the edit has less to anchor to — crop to the subject you want changed and make it as large as possible.
- What to reference: the first image should be the photo you want to edit; if you have a style reference (a colour palette or mood board), that’s a fine second input.
Step-by-Step: Edit with a Reference Photo
Open the Image Generator and switch models:
Edit an image with a reference photo for free →
6 Real‑World Edit Prompts That Work
Good reference‑edit prompts describe the change, not the whole scene. Copy, paste, and tweak:
1. Colour swap
Change the woman’s red dress to emerald green, keep lighting and background unchanged
2. Background replacement
Replace the plain wall behind the plant with a sunlit garden, keep the plant exactly the same
3. Style/lighting tweak
Turn this into a cinematic noir scene: low single light, strong shadows, film grain, keep faces the same
4. Add a removed object
Add a steaming coffee cup onto the desk where the empty space is, photo‑realistic, natural lighting
5. Season change
Make the same street look like winter: light snow on the ground, bare trees, same perspective and camera angle
6. Style transfer via second reference
Repaint the house in the Van Gogh brushstroke style, matching the reference painting’s palette, keep the shape of the house
Edit vs Generate vs Upscale — When to Use Which
| Goal | Use this |
|---|---|
| I have a photo and want to change part of it | Qwen‑Image‑Edit‑2511 (reference edit) |
| I have a description only, want a new image | Z‑Image‑Turbo (draft) or Qwen‑Image‑2512 (final) |
| My image is small/pixelated and I need it bigger & sharper | AI upscaling |
| I want my photo redrawn in a new art style | re‑Imagine style transfer |
Tips for Predictable Edits
- Crop the reference to the subject. Less background = less for the model to “re‑imagine”. Fill the frame with what you want changed.
- State what to keep. “keep the face the same”, “unchanged” — anchoring language prevents unwanted drift.
- One change per prompt. “Change the dress colour and the background” often loses both; do colour first, background second.
- Use the second reference for style. Upload a mood‑board/photo as the second image to nail palettes and lighting.
- Expect ~3 minutes. Diffusion takes longer than plain generation — queue ahead for batches.
Frequently Asked Questions
Can I really edit a photo with AI for free?
Yes. Qwen‑Image‑Edit‑2511 costs €0.12/image and runs on the free daily quota (€0.30/day) — so the free tier covers a few edits per day. No sign‑up, no watermark.
How many reference photos can I upload?
Up to 2. Use the first for the image you’re editing; optionally a second as a style/mood reference to guide colour and lighting.
Is this like Photoshop’s “Generative Fill”?
Conceptually, yes — both describe a change in words and let AI redraw it. The difference here is no install: the model runs server‑side and returns a PNG to your gallery.
What if the edit changes the face I wanted to keep?
Be explicit in the prompt: “keep this person’s face exactly the same”, and crop the reference so the face is centred and large. You can re‑run with a refined prompt — each attempt is just €0.12.
Are my photos stored or used to train models?
No. Reference images and results are matched to your random cookie, kept on this German host for 3 hours, then auto‑deleted. Never used for training, never shared.
Ready to Edit Your Image?
“Edit image with reference photo” shouldn’t mean installing a 5 GB AI toolkit. With Qwen‑Image‑Edit‑2511 you get reference‑guided editing in ~3 minutes, free on the daily quota — colour swaps, background changes, and style transfers included.
Edit an image with a reference photo for free →
No account required · Free daily quota · No watermark · Images auto‑deleted
For plain text‑to‑image, see the text‑to‑image guide, and for full‑quality portraits the Qwen Image guide. Questions? Discord.