Upload a photo. Describe the change. Skrrol's image-to-image rewrites the image in place — backgrounds, styles, objects, lighting — and hands the result to a real editor.
Image-to-image is the practical workhorse of AI image work. Instead of starting from nothing, you start from a photo you already have — a product shot, a portrait, a screenshot, an old generation — and tell the model what to change. "Replace the background with a city skyline at dusk." "Turn this into an oil painting." "Make it look like it was shot on 35mm film." "Remove the person on the left." The model rewrites the image in place.
Skrrol AI runs image-to-image through OpenAI's GPT Image, the model behind DALL·E's successor, which handles instruction-driven edits cleanly. You can do four broad jobs with it: style transfer (turn a photo into an illustration, painting, or different rendering style), object edits (add, remove, swap, recolour specific things in the scene), restaging (move the subject to a different background or lighting setup while preserving identity), and retouching (clean up small flaws, brighten, even out skin, polish the frame).
For production teams, this replaces a sizeable chunk of standard photo-editing work. Product variants — same product, different scene — used to mean a new shoot or a careful manual retouching session; now it's one prompt. Brand-consistent moodboards built from a single hero shot used to take an afternoon; now they take ten minutes. Restoration of old photos, adding people to a group shot, fixing eye direction in portraits — image-to-image handles them.
Once the rewrite lands, the Skrrol editor opens on it for finishing work — crop, color-grade, mask, layer, type. If a single edit didn't quite land, re-prompt and try again with a refined instruction. The Free tier is enough to try the workflow on a few images; paid tiers unlock the volume needed to produce variant packs at agency speed.
Plain-English prompts run on top of an uploaded photo. The model interprets and rewrites in place.
Turn a photo into an illustration, painting, comic, watercolour, 3D render, or any other prompted style.
Remove a person from a group shot, swap a product colour, add a prop into the scene — all from a prompt.
Keep the subject and rewrite the environment — different background, different lighting, different setting.
Clean small flaws, even skin, balance light, polish the frame. Fast alternative to a manual retouch pass.
After the rewrite, crop, color-grade, mask, and layer in the Skrrol editor. No tool-hopping.
Result: The same portrait restaged outdoors with believable matching light — useful for headshots, ad creative, and avatar generation.
Result: An on-style illustration that preserves the dog's identity but reads as artwork — useful for prints, cards, and gift products.
Result: A clean version of the photo without the distraction — saves a careful manual content-aware fill.
Result: A new product variant generated without re-shooting — combine multiple variants into a marketplace listing.
Sign in, click Generate, pick the Image tab, and switch to Image-to-Image mode.
Drop in any image from your device — JPEG, PNG, or HEIC. The image stays on your device until generation runs.
Be specific about what to change and what to preserve. "Replace the background" is fine; "replace the background with a sunlit park, keep her face lighting unchanged" is better.
Hit Generate. The model returns the rewritten image in seconds. Re-roll if needed with a refined prompt.
Open the result in Skrrol — crop, color-grade, mask, and layer. Stack multiple edits if needed.
Export the final image, or feed it into image-to-video to animate it into a clip.
Skrrol AI uses VL credits across all generators — image, video, voice, and music. The same credit pool applies; heavier modalities (video) use more credits per generation than lighter ones (image, voice). Choose a plan and use credits across any generator.
Starter VL credits to test prompts on entry-tier image models. Watermark on free outputs. Great for finding your style before subscribing.
8000 VL credits — hundreds of mid-tier image generations per month. Suitable for thumbnails, blog hero art, and social posts at a steady cadence.
17000 VL credits — heavier volume for moodboards, ad creative iteration, and product photography variants.
35000 VL credits — agency-grade volume for brand campaigns, illustration libraries, and large-format renders.
GPT Image is good at preserving the subject when the prompt asks it to. For best results, name what should stay ("keep the face unchanged") and what should change ("replace the background"). For extreme transformations, expect more drift — re-roll if the identity moves too far.
Standard photo sizes work fine. Very large images may be downscaled by the model before generation; the output is returned at the model's native resolution.
Yes — chain instructions in a single prompt ("remove the sign, brighten the sky, add a sunset"). For best control, do one big change per generation and chain them.
The source goes to the underlying model provider so the model can run. Skrrol does not retain it on a content cloud — outputs land in your local project.
Yes, with care. It's good for general polish — even skin, balance light, fix small flaws. For high-stakes portraiture, manual retouching is still safer because models can drift on identity.
Roughly the same — image generation is the cheapest modality on Skrrol regardless of input mode.
Every generation opens directly in the Skrrol editor. These features are particularly useful as the next step after a image-to-image ai — edit photos with a prompt run.
Skrrol AI runs every generator next to a full pro editor. Your work stays on your device. Start free.