FLUX.2 AI Image Generator for Reference-Led Character and Product Edits
Start from a source image, preserve the visual anchor, and change one part of the scene at a time.
FLUX.2 is the lowest-cost reference-capable image model in this catalog and is eligible for the first free generation. It is the practical starting point for character anchors, product edits, and controlled visual variations; use Recraft V4 when the deliverable needs editable SVG geometry instead.
FLUX.2 earns its place in a production workflow through editing, not just one-shot rendering. Upload the approved character or product, state what should change, and judge whether everything else survived. Specific prompts outperform vague art direction.
How to get consistent work from FLUX.2
Approve a clean anchor
Choose a reference that clearly shows the face, silhouette, materials, and palette. Even light and an uncluttered background give the model better identity evidence than a dramatic image with hidden details.
For a character, lock the face and costume before changing pose or scene. For a product, verify package proportions, closures, logos, and label placement before generating lifestyle variations.
Describe only the edit
FLUX.2's catalog prompt tip is unusually important: say what should change and leave approved details alone. “Same character; concerned expression; preserve hair, jacket, crop, and lighting” is easier to review than rewriting the whole portrait.
Change expression, pose, wardrobe, and environment in separate passes. When drift appears, a one-variable workflow makes the cause visible.
Hand off by output type
Keep FLUX.2 for raster stills and reference continuity. Send flat icon and callout needs to Recraft V4, quoted title text to Ideogram 3.0, and approved stills that need motion to Kling or Wan.
The AI companion, product photo, and character sheet guides show those workflows in context.
What you can create
- Character anchor portrait
- Establish the face, hair, costume, and palette that later expressions and scenes will reference.
- Expression variation
- Reuse the anchor and change brows, eyelids, and mouth while preserving head angle and clothing.
- Product background edit
- Replace the setting or lighting around an approved SKU, then compare packaging and geometry with the source.
- Concept-art iteration
- Explore one silhouette, material, prop, or environment change at a time instead of rebuilding the entire brief.
Reference input reduces drift; it does not eliminate it
Inspect facial proportions, asymmetrical details, hands, text, and costume closures after every generation. A plausible variant can still be a different character or product.
FLUX.2 rewards specificity. Vague requests force the model to make more design decisions and make accepted details less stable.
When to choose another model
Use the same prompt in the comparison studio when the deliverable crosses model specialties.
Nano Banana Pro
8 credits · Google
Use Nano Banana Pro when the edit has many explicit conditions or short text.
Reference image supported.
Recraft V4
6 credits · Recraft
Use Recraft for editable SVG icons, badges, and flat callout graphics.
Text prompt only.
GPT Image 2
12 credits · OpenAI
Use GPT Image 2 for complex photoreal scenes and multi-object composition.
Reference image supported.
Best workflows for this model
Frequently asked questions
Does FLUX.2 accept reference images?
Yes. The current catalog supports image input and up to four outputs per request.
Can FLUX.2 keep a character perfectly identical?
No image generator guarantees pixel-identical continuity. A clean reference and one-variable edits improve consistency, but every result still needs review.
How much does FLUX.2 cost?
FLUX.2 costs 6 credits per request and is eligible for the first free generation.
When should I use Recraft instead?
Use Recraft V4 when the required output is editable vector art such as icons, emotes, badges, or diagrams.
More image models