6 min read
How to Edit AI Photos Without Losing Your Character
The worst way to fix an almost-right AI photo is to regenerate from scratch and hope. The canvas's edit tools work like a retoucher instead: describe the one thing to change — the lighting, the background, the outfit color, the expression — and everything else, including your character's face, stays put. Edits land beside their source on the board, so you can compare versions and keep the winner.
step by step
Open the edit tool
Hover an image and choose Edit (or use /edit in chat with the image selected). The edit modal takes a plain-language instruction and offers a model picker if you want a specific engine for the job.
Describe one change
Be surgical: "make it nighttime", "change the background to a beach", "make the outfit black", "add a soft smile". Quick-edit chips in the modal cover the most common fixes in one tap.
Compare on the canvas
The edited version appears next to the original, linked to it. Generate several competing edits from the same source, line them up, and delete the losers.
Chain edits
An edit is a full image node — edit it again, animate it, or select it as a reference. Complex transformations work best as a chain of small verified steps rather than one giant instruction.
What edits are good at
Instruction-based editing shines when the change is local or global-but-simple: relighting (day to night, golden hour), background swaps, clothing color, weather, adding or removing objects, and expression adjustments. It preserves composition and identity because it starts from your pixels rather than re-imagining the scene. When you want a fundamentally different composition or pose, that's a new generation with the original selected as a reference — not an edit.
The rest of the retouching kit
Editing is one of several per-image tools on the node toolbar and slash menu.
Remove background
One click produces a transparent PNG beside the original — up to four images at once. Perfect for thumbnails, stickers and composites.
Regenerate
Re-rolls the same prompt and settings for a fresh take when the concept is right but the execution isn't.
Copy prompt
Pulls any image's prompt back into chat, ready to tweak — the fastest way to make a deliberate variation.
/describe
The agent looks at a selected image and writes a reusable shot prompt from it — costing no credits. Great for reverse-engineering the look of your best shot.
Editing without losing the person
Because edits start from an image that already contains your character, likeness survives almost any instruction — you can change wardrobe, weather and time of day freely. The riskiest edits are the ones that touch the face itself; keep those minimal ("soft smile", not "different expression entirely") and prefer regenerating with a strong reference when a face needs major work.
quick tips
- One instruction per edit. "Nighttime, new background, red dress" fights itself — chain three small edits instead.
- Keep originals until a series ships; edits are linked to their source, and you'll often want to branch differently later.
- Use remove-background outputs as props or composites in other tools — transparent PNGs travel well.
- If an edit ignores you, rephrase toward the concrete: "make the sky dark blue with stars" beats "more atmospheric".
questions, answered
Does editing change my original image?+
No — edits always produce a new node beside the source. The original stays untouched until you delete it yourself.
Can I edit photos I upload, not just generated ones?+
Yes. Anything that lands on the canvas as an image node can be edited, background-removed, animated or referenced.
Which model should I pick for edits?+
The default handles most instructions well. If a particular edit keeps missing, switching models in the edit modal often resolves it — different engines have different strengths.
Can I edit videos the same way?+
Trim, mute and text overlays are available on every video. Generative video restyling ("make this anime") uses the edit_video tool on Pro and Studio plans.
keep learning
consistent characters →
Character consistency is the whole point of an AI photo studio. Here's how selfielab keeps one face across every shot — uploads, personas, styling, and reference images.
photo to video →
Animate any canvas image into video, direct the camera with motion presets, extend clips, stitch them together, and screenshot the perfect frame back into a photo.
shot settings →
Framing, lenses, lighting, color grades and camera styles: use selfielab's shot settings chip and presets like iPhone Candid or Studio Editorial to shoot like a photographer.