FLUX.2
Our shared brief is a woman in a crimson coat with an orange cat at a rain-soaked train platform at blue hour. The goal is to keep that recognizable scene while changing the instructions for a still image, an image edit, or a moving shot. Compare the subject, setting, action, camera, light and what must remain consistent. Promptlyfied writes prompt text; the selected image or video service produces the media.
When to use this workflow
Ask whether you need a fresh image, a controlled edit, or a composition assembled from references. That choice changes the prompt more than an impressive adjective does. A new image needs subject, setting, composition and light; an edit needs a source image plus a precise change and preservation instructions.
FLUX.2: combine text and references
Black Forest Labs describes FLUX.2 as an image generation and editing family with multi-reference workflows. When references are available, refer to them by role: person from image 1, coat from image 2, street lighting from image 3. Write the desired final scene as a clear whole, then identify details to preserve. Check the specific FLUX.2 variant and product for supported inputs.
Example 1: the rainy platform
See how the same rainy train-platform concept changes when the goal is a still image, an edit, or a moving shot.
Create a wide editorial photograph: a woman in a crimson coat and an orange cat on a rain-wet train platform at blue hour, warm carriage lights reflected in puddles. If image 1 is supplied, preserve the woman’s identity and outfit while changing only the background.
Why the prompt is written this way
Separate a new-image description from an edit with a supplied reference. Check your FLUX.2 variant’s controls.
Example 2: a different scene
Use image 1 as the product reference. Create a studio photograph of the same ceramic teapot on a dark green stone table. Preserve its handle, glaze pattern and proportions; change only the setting and light to a soft window glow from camera left. Leave negative space on the right.
How to adapt the second example
The second prompt is an edit, so the attached product photo is essential. Replace “image 1” with the reference role used by your chosen interface and list only the details that must survive.
If the result misses the brief
Use the same subject and output shape when comparing models, then adapt any model-specific parameters. Inspect composition, identity, text rendering and editing precision separately. The most effective wording depends on the source images and model version; a universal “best model” claim would mislead readers.
Start with “Use the attached image as the source.” Follow with one change, preservation constraints, framing, and lighting. Review the first output before adding more detail. Strong prompts describe the target image, not a string of generic quality tags.
Try the matching prompt tool ↗Reviewed 2026-10-12