Venturetwins reports nine FLUX 3 edits without visible degradation

• One Chloe ad survived nine successive FLUX 3 edits without visible degradation, according to venturetwins.
• FLUX 3 supports bounding-box layouts, up to ten references and 4K output, according to BFL's launch thread.
• Magnific now offers the model with element placement and targeted detail editing, as described in its announcement.

The Chloe makeover reaches all the way to the logo. Black Forest Labs' layout controls use a 0-to-1000 coordinate grid, while a creator's three-model comparison targets tiny marble columns with camouflage.

venturetwins used an old Chloe ad to test whether FLUX 3 could replace its styling and set dressing while maintaining the structure. The changes covered:

• Dresses
• Couch
• Wallpaper
• Makeup
• Logo

“Nine turns later and no artifacts!” they reported.

BFL's launch thread describes four controls:

• Bounding boxes: placement controls for compositions from scratch and precise edits.
• References: up to ten source images combined into one composition.
• Local editing: targeted changes that BFL says leave other pixels unchanged.
• Resolution: generation at up to 4K.

Bounding boxes are a welcome bit of manual control for layout obsessives.

ai_artworkgen used Midjourney originals to compare Midjourney, Ideogram 4.5 and FLUX 3, describing an edit as “quick” in a reply. Their kimono test used three requests:

1. “Change her kimono to pink silk with white flowers, and a custard yellow bow.”
2. “Show me a macro closeup of the subject's face.”
3. “Show me a macro closeup of the subject's hands on her kimono.”

Their assessment in the same thread:

• Midjourney: changed the garment color and bow, but missed the white flowers and exact yellow shade.
• Ideogram 4.5: fulfilled the clothing request, though the close-ups' lighting felt off.
• FLUX 3: fulfilled the clothing request and produced stronger close-up lighting and contextual consistency, in their view.

For a smaller target, ai_artworkgen requested: “Change some of the detailing on the marble columns to camo print.” They reported that only FLUX confined its changes to the columns.

venturetwins also tested Ideogram's new editing model on an Olipop ad, exploring variations and localization. They reported a series of precise changes without warping or artifacts.

In their own tests, ai_artworkgen praised Ideogram's iterative editing, including some iterations with two changes at once. Magnific offers Ideogram 4.5 on all plans across desktop, web and Magnific MCP.

Runware's FLUX 3 editing guide documents a mask-free, single-reference workflow:

• Source: the image goes in inputs.referenceImages.
• Instruction: positivePrompt describes the change.
• Result: the response returns an imageURL.
• Dimensions: output follows the requested width and height, rather than automatically matching the source image.

Its example instructions have three parts:

1. The specific target, such as “the mustard yellow corduroy overshirt.”
2. The finished state, such as “a slate blue one in the same corduroy.”
3. The elements to preserve, including the face, other garments, framing and lighting.

In a separate cat-video experiment, venturetwins described their tooling as a fal API key plus a custom Codex setup.

FLUX 3 Image is available through Magnific.

BFL describes three separate access terms in its launch thread:

• BFL API: 50% off until October 8.
• Commercial weights: a license for companies to fine-tune the model and deploy it on their own infrastructure.
• Open weights: a version planned for the coming weeks.