You have four recipe step photos and want one illustrated look without losing what changes between the steps. This CapCut test gives you four source-and-result pairs and a record of the ingredient counts, visible preparation states, and placements that survived the restyle.
The photographic source examples were generated in CapCut for this demonstration; they do not document a real cooking session. One 2048 × 1152 contact sheet was cropped into four 1024 × 576 source files outside CapCut, without repainting their pixels. Browning here is a visible illustration cue, not evidence of doneness or food safety. This walkthrough covers still images, rather than recipe cards or cooking videos.
Follow the change from two tomatoes to four pieces
Before using an AI image transformer, write down what each source actually shows. In this set, step 01 has two whole tomatoes on a cutting board. Step 02 has four separate tomato pieces with exposed seeds and bright cut faces. The count changes because the cutting step changes the visible units.
Step 03 keeps four pieces, now with brown patches on their cut faces, inside a black pan. Step 04 moves those pieces onto two toast slices on a white plate, with two pieces on each slice. The useful check is whether each illustration matches its own source and still connects to the next step.
Both early steps passed: two whole tomatoes remain whole, while the next image retains four separate cut faces. Brush texture changes; the visible cutting step remains distinct.
For your own sequence, number the source files before uploading them. Record visible pieces rather than inferring weight or hidden ingredients from a photograph. A top view may show a round cut face clearly while leaving the thickness uncertain.
Keep one gouache brief across the recipe step photos
1. Start from CapCut’s image style transfer page and open the image workspace available to your account. In the web editor tested on October 21, 2026, the panel was called Design with AI. The model selector showed GPT Image 2 as the single selected option. That records the visible selector; the underlying model was not independently confirmed.
2. Use New chat, then Upload image, and attach the first source. Check the attachment chip before sending. In this editor, a selected canvas image could also appear as a reference; pressing Escape to deselect it before uploading helped keep the intended source as the only attachment.
Two actual UI crops show the single source attachment and selected model. Amber outlines mark the controls used; account information is excluded.
3. Send this same brief for each source. Its style wording stays fixed so the comparison concerns the recipe steps rather than four different visual treatments.
Restyle the attached recipe step photograph as a flat gouache illustration with warm muted colors, visible brush texture, simplified shapes and dark outlines. Keep the same food and arrangement. Preserve the original stage of preparation. No new ingredients, labels, text or decorations. Keep the original 16:9 landscape framing. Return one image only.
4. Inspect the first result before continuing. Here, the two tomatoes remained whole and separate, so the same prompt was sent in three further new chats, each with its own original step file. The four restyled images in this run were 2048 × 1152. Access and credit requirements should be checked in your current account before submitting.
Separate browning from the warm palette
Warm paint colors spread across the board, food, and background in these results. That warmth alone does not tell you whether the preparation stage survived. Compare the cut faces in steps 02 and 03: the first keeps bright seed-filled centers, while the pan image retains darker brown marks across the tomato surfaces.
Enlarged crops from the same sources and final exports keep the surface comparison visible. The brown patches are stronger in the pan step than in the cutting-board step.
Next, trace the container boundary. In step 03, the four pieces stay inside the black pan. In step 04, the tomatoes sit on the bread rather than appearing as spare pieces on the plate. Count each toast separately: a total of four would still conceal a changed arrangement if one slice had three and the other had one.
The later pair retains four pieces in the pan, then two on each toast. No extra garnish or spare tomato piece appeared in these final exports.
If your result adds an ingredient, loses a piece, or removes the visible distinction between stages, hold that image out of the set. Return to its original source for a focused retry and name the changed fact alongside the unchanged style brief. For a wrong topping count, that might mean specifying two tomato pieces on each of two toast slices. This is a conditional correction to try, not a second prompt tested in this run: the four first-pass results met the recorded checks, so no content retry was needed.
Read the four illustrations in source order
Place the accepted results in the original order and read the changes without relying on captions: whole tomatoes become cut pieces, the cut faces gain brown marks in the pan, and those pieces move onto toast. The small differences in board texture and illustrated outlines did not remove those transitions in this set.
The four accepted illustrations appear in source order, read across each row. These are native still exports; the HTML version uses a locally assembled source/result review animation.
For the download, select the intended image in Layers and wait for it to finish appearing on the canvas. Open Download, confirm the selected filename, and export as PNG. The final review here used the four restyled images from one download batch after reopening the saved project. Keep your original source files alongside them, named by step, so a later revision can be checked against the photographed state rather than an earlier illustration.