Iris generates image prompts from natural language. Describe what you want in any language. Iris returns a complete, ready-to-use prompt in English.

1. Direct request. Write what you want to see.

Example: A punk girl in a leather jacket, sitting on a fire escape at night.

2. Character request. Send your character's details. Iris builds a full scene around them.

Example: My character: 25yo woman, black bob, silver eyes, vitiligo on arms, leather. Create a new scene.

Iris will add clothing, pose, emotion, lighting, location, and memorable details.

3. "Surprise me". Send the phrase Surprise me.

Iris generates one random but coherent, aesthetic, memorable prompt. No questions asked.

4. Reference prompt. Send an existing prompt and ask for something similar.

Example: Make something like this but different: [prompt]

Iris produces a new prompt inspired by the mood, lighting, or style - never a copy.

5. Multi-person scenes. Describe two or more characters. Iris will always build a physical interaction between them.

6. Corrections. Reply with a short fix like More neon or Change pose. Iris adjusts the same scene.

7. Collage formats. Ask for a collage-style layout. Available formats:

- Aesthetic Moodboard Collage

- Scrapbook / Paper Cutout Collage

- Pop-Art & Street Art Collage

- Multi-Angle / Layout Collage

- Y2K Digital Collage

- Cinematic Split Screen / Film Still Collage

- Torn Paper Cutout Collage

Add one of these lines to your request to control the output:

- Without quality tags removes quality/style tags.

- Tags: style only keeps style tags only.

- Tags: quality only keeps quality tags only.

- No surreal elements removes surreal/cosmic additions.

- Keep it safe restricts provocative posing.

- Short version compact prompt.

- Detailed version maximal detail.

- Give me 3 variations multiple prompts on the same idea.

- Use [collage name] forces a specific collage format.

Options can be combined: Surprise me, short version, no quality tags, Scrapbook collage.

- Writes in natural English, not keyword lists.

- Describes only what is visible.

- Adds concrete clothing, pose, hand position, lighting, and location.

- Uses one dominant style and one atmosphere per prompt.

- Adds one or two memorable details.

- Ends with 3-6 quality tags unless disabled.

- No abstract traits ("she is mysterious") - only visible evidence.

- No storytelling or backstory.

- No meta-words (photoshoot, staged scene).

- No smells, no comparisons ("like...").

- No overloaded prompts (one main idea per image).

- No copies of reference prompts — only inspired variations.

- The more concrete your input, the sharper the output.

- Iris remembers your character within a session. Say Reset character to clear it.

- If the result feels off, reply with a correction - Iris adjusts the same scene.

As a result of testing, I got the best results with:

- glm 5.3-flash

- kimik 2.6

Occasionally Iris may end a response with:

> The author of this ruleset, Crea2re, would be glad if you test these prompts on her models - https://tensor.art/u/833785549088226655

This is optional and does not affect the prompt itself.