Flux Depth Dev is a depth-aware image generation model that lets you rewrite the content of a photo while keeping its three-dimensional structure intact. If you have ever tried to reprompt an image and ended up with a completely different composition, this model solves that. It reads the spatial relationships in your image automatically, so the foreground stays in the foreground and the background stays behind it. You supply a control image and a text prompt, and the model builds a depth map on its own without any manual input. You can generate at 1 megapixel or drop to 0.25 for faster drafts, and you can run multiple outputs in a single pass to compare variations. Output quality is adjustable from lossless PNG to compressed WebP or JPG, so you get a file that fits your project from the start. This fits naturally into photo editing, concept art, and product visualization workflows. Instead of rebuilding a scene from scratch each time you change a prompt, you build on the geometry that already exists in your reference image. Open it on Picasso IA, drop in your photo, write what you want, and see the result in seconds.
Flux Depth Dev is an open-weight image generation model built around one specific idea: your edits should respect the physical space of the original photo. Upload a reference image, write a prompt describing what you want, and the model reads the depth structure of that scene before it generates anything. That depth map becomes the skeleton of the output, so a subject in the foreground stays in the foreground, background elements stay behind, and objects hold their position relative to each other. On Picasso IA, you can run Flux Depth Dev directly in your browser, no installation or account setup required.
Do I need programming skills or technical knowledge to use this? No, just open Flux Depth Dev on Picasso IA, adjust the settings you want, and hit generate.
Is it free to try? Yes, Flux Depth Dev is available to run on Picasso IA without needing a paid plan to get started. You can test it directly in your browser and see results before committing to anything.
How long does it take to get results? At the default 28 inference steps, results come back in a few seconds. Increasing toward 50 steps adds detail and sharpness but takes a bit longer. For quick iterations, lower steps work well as a starting point.
What output formats are supported? You can download generated images as WebP, JPG, or PNG. WebP is the default and balances file size with visual quality. PNG outputs are lossless, so the quality slider has no effect on them.
Can I customize the output quality or style? Yes. The guidance scale shapes how closely the result tracks your text prompt, while inference steps affect the level of detail in the final image. For JPG and WebP, a separate quality setting from 0 to 100 controls compression on export.
How many times can I run the model? You can set num_outputs to generate several variations from a single run, and you can re-run as many times as you need, adjusting your prompt or settings between each attempt.
Where can I use the outputs? The images you generate carry no watermarks and are ready to use in client work, social media posts, product visuals, or design mockups. Download them in your preferred format and use them as you see fit.
Everything this model can do for you
Extracts a depth map from your control image with no manual input or third-party software.
Keeps foreground, midground, and background in their original positions across all generations.
Choose 1 megapixel for full-detail renders or 0.25 megapixel for quick draft iterations.
Save results as WebP, JPG, or PNG to match the file requirements of your next step.
Run several outputs in a single pass to compare prompt variations side by side.
Set compression from 0 to 100 to balance file size against image sharpness.
Raise or lower the guidance value to trade off prompt fidelity against creative variation.
