Image to Flux Prompt Generator

Why Flux prompts are completely different

Flux was architected around the T5 text encoder, which understands full sentences the way GPT-4 does. This is why Midjourney and SDXL habits actively hurt Flux output. The tool below generates prompts in Flux's native 3-part structure: subject & scene → photographic detail → mood & style.

1. 1. Subject & scene (first sentence)

   A woman in a red silk dress standing barefoot on a sandy beach at sunset

   Flux weights earlier tokens heavily. Lead with who, what, where in one coherent sentence.

2. 2. Photographic specifics

   shot on Hasselblad X2D, 85mm lens, shallow depth of field, soft bokeh across the water

   Flux's T5 encoder recognizes real camera/lens names and applies their aesthetic signatures — sensor size, dynamic range, color science.

3. 3. Light interaction (not conditions)

   warm golden sunset light streaming low across the scene, casting long shadows, dust particles visible in the beam

   Instead of 'golden hour', describe what the light *does*. Direction, quality, interaction with surfaces — this is where Flux outperforms everything else.

4. 4. Mood / style (optional closer)

   cinematic, moody, editorial fashion photography

   Add context flags at the END — Flux uses them as style anchors after the scene is set. Putting them first wastes attention on abstraction.

5. 5. What NOT to include

   No --ar 16:9. No (beautiful:1.4). No 8k, masterpiece, best quality tags. No negative prompt.

   All of these are Midjourney/SDXL syntax. Flux ignores them at best, prints them as literal text at worst (especially Flux Dev with --flags).

Real image, real prompt

The prompt below is the raw output of this tool on the image shown — not hand-written samples.

A man with dark hair wearing a cap, facing away, sitting at a table, looking at a laptop screen. Cafe interior: gray concrete floor, white walls, black chalkboard menu with white text, wooden bar with bottles, white chairs with orange pillows, large green potted plant, hanging pendant lights, other patrons in the background. Soft diffused light, warm color temperature, shadows are soft.

Things most beginners miss

1. 1. Treat Flux like you're briefing a photographer

   Write prompts as if you're describing the shot you want to a human with a camera. Tell them the subject, the setting, the light, the feel. This maps directly to how Flux's T5 encoder processes your input.

2. 2. Real camera names unlock real aesthetics

   'shot on Hasselblad X2D' produces different output than 'shot on iPhone 15 Pro' — Flux knows these sensors' color science. Specific gear > generic 'professional camera'.

3. 3. No negative prompts — describe positively instead

   Flux Dev/Schnell doesn't support negative prompts. If you want a clean background, don't negative-prompt 'cluttered'; write 'minimal, empty background' in the positive prompt.

4. 4. Flux Dev hates 'white background'

   A known Flux Dev quirk: plain 'white background' produces flat, unclear outputs. Add environment detail instead: 'seamless studio backdrop under soft key light' or 'pure white cyclorama'. Flux Pro/Schnell don't have this issue.

5. 5. Guidance 3.0–3.8 for Flux Dev, default for Pro

   If you're running Flux Dev locally: steps 24–32, guidance 3.0–3.8, resolution 768–1024. Flux 1.1 Pro's API handles these automatically — you only pass width/height and optional seed.

Flux vs other image models

Frequently asked questions

Other model guides

• Image to Midjourney Prompt— Prefer stylized aesthetics over photorealism, or want --sref for style consistency? Midjourney V7 is still the style king.
• Image to SDXL Prompt— Want to run locally on 8GB VRAM (not 12GB+)? SDXL is the practical open-weights choice below Flux Dev's hardware requirements.