Prompt Tips
Updated May 2026
This page covers how to write prompts that actually work for three things: generating 3D models from text, generating images, and editing existing images. No theory, just practical tips and real examples.
Text to 3D Prompts
When you write a prompt for Text to 3D, the AI needs to understand what object to build, what it looks like, and what style you want. Keep it between 5 and 20 words for most cases. Be specific but don’t overload it.
The formula
Object + key features + material/style + constraints
Prompts that work
| Prompt | What you get | Good for |
|---|---|---|
| ”medieval castle with stone walls and a tall tower” | Detailed architectural model | Environment assets |
| ”cute chibi anime cat girl, big eyes, full body, pastel colors” | Stylized character | Game characters, figurines |
| ”futuristic sci-fi pistol, metallic, detailed, hard surface” | Clean weapon model | Game props |
| ”red sports car, glossy paint, side view, realistic” | Vehicle with good proportions | Product visualization |
| ”wooden stool, three legs, worn oak, subtle wood grain, one solid piece” | Clean furniture model | Interior design |
| ”fantasy potion bottle, chunky glass, wide base, glowing liquid inside” | Stylized prop | RPG game assets |
| ”sci-fi helmet, smooth curved shell, dark matte polymer, symmetrical” | Hard surface model | Character equipment |
| ”cute chubby robot mascot, round body, simple shapes, friendly face” | Cartoon character | Branding, mascots |
What makes prompts better
- Name the material - “wooden”, “metallic”, “glass”, “plastic”, “stone” gives the AI clear texture direction
- Mention the style - “realistic”, “stylized”, “low-poly”, “cartoon”, “photorealistic” sets the aesthetic
- Say “one solid piece” - prevents the AI from generating floating parts or disconnected geometry
- Add “clean geometry” - helps with printing and game engine compatibility
- Specify view - “front view”, “side view”, “three-quarter view” for more predictable results
- Use real names - “wingback chair” works better than “chair with wings on the sides”
What to avoid
- Too vague - “a cool thing” or “something nice” gives the AI nothing to work with
- Too long - paragraphs of text can confuse the model. 5-20 words is the sweet spot
- Contradictions - “realistic cartoon” or “simple with extreme detail” sends mixed signals
- Multiple objects - “a sword and a shield and a helmet” usually produces a messy result. Generate one object at a time
- Scene descriptions - “a knight standing in a forest next to a dragon” is a scene, not a single 3D model. Focus on one object
Model-specific tips
| Model | Works best with | Tip |
|---|---|---|
| Prism 3.1 | Detailed objects, realistic style | Add “PBR texture” for better materials |
| Prism Turbo | Quick concepts, simple objects | Keep prompts shorter (5-10 words) |
| Hunyuan 3.1 Pro | Organic shapes, characters | Try the “Geometry” mode for untextured sculpts |
| Meshy 6 | Characters, creatures | Use “full body” and specify a pose |
| Forge Gen-2 | Detailed props, hard surface | Mention specific materials and surface finish |
| Tripo P1 | Game-ready assets | Add “game ready” or “game asset” to the prompt |
Image Generation Prompts
When generating images in Image Studio, different AI models need different prompting styles. But all of them respond well to clear, descriptive language.
The formula
Subject + action/pose + style + context/lighting
Prompts that work
| Prompt | Model | What you get |
|---|---|---|
| ”stylized 3D cartoon owl, soft lighting, clean white background” | ImageGen 4 | Clean reference for Image to 3D |
| ”professional product photo of a leather watch, studio lighting, white background” | Flux 2 | Product photography |
| ”anime warrior girl with blue hair and katana, full body, cel shaded” | ImageGen 4 | Character concept art |
| ”LEGO minifigure of an orc warrior, plastic material, studio lighting” | ImageGen 4 | Stylized figurine for 3D conversion |
| ”architectural interior, modern kitchen, natural window light, wide angle” | Flux 2 | Photorealistic interior |
| ”pixel art game character, knight with sword, 16-bit style, transparent background” | Flux 2 | Game sprite |
Model-specific advice
Flux 2 - Write prompts like natural sentences, not keyword lists. Flux understands flowing descriptions better than comma-separated tags. Put the most important element first. No negative prompts needed. For photorealism, mention camera models (“shot on Fujifilm X-T5”) and lighting setups.
ImageGen 4 - Great at following complex instructions with minimal prompting. Produces photorealistic results with simple descriptions. Good at text rendering in images.
GPT-Image-1 - Excellent at following detailed instructions. Works well with structured prompts (scene, subject, details, constraints). Best for text-in-image and complex compositions.
For Image to 3D conversion
If you’re generating an image to convert to a 3D model afterward, add these to your prompt:
- “clean white background” or “studio lighting” - makes the AI focus on the object
- “full body” or “complete object visible” - prevents cropping
- “single object, centered” - avoids confusion about what to convert
- “3D render style” or “stylized 3D” - produces images that convert to 3D better than flat illustrations
Image Editing Prompts
When editing images in Image Studio (using GPT-Image-2 Edit, Gemini Edit, or Nano Banana), the key principle is: say what to change AND what to keep.
The formula
What to change + what to preserve + style/quality constraints
Prompts that work
| Prompt | What it does |
|---|---|
| ”turn this person into a LEGO minifigure, plastic material, clean background. Keep the pose and general appearance.” | Style transfer while preserving identity |
| ”remove the background. Keep the subject exactly as is.” | Clean background removal |
| ”change the lighting to warm sunset. Keep everything else the same.” | Relighting |
| ”turn into a stylized 3D figurine, full body, clean background” | Convert photo to 3D-ready reference |
| ”make it Ghibli style. Preserve the composition and subject position.” | Art style transfer |
| ”replace the jacket with a navy windbreaker. Do not change face, pose, lighting, or background.” | Surgical clothing swap |
Key principles
- Be specific about what stays the same - “keep the face, pose, lighting, and background unchanged” prevents drift
- One change at a time - don’t try to change background, clothing, and style in one prompt. Do it in rounds.
- Describe the target, not the process - “warm sunset lighting” works better than “make the image warmer”
- For text in images - put the exact text in quotes: ‘write “SALE 50% OFF” in bold white letters, top center’
- Match lighting - if the original has warm light, don’t ask for cold light in the edit or the result will look composited
Model comparison for editing
| Model | Best for | Notes |
|---|---|---|
| GPT-Image-2 Edit | Precise edits, text rendering, photorealism | Most capable overall, best at following complex instructions |
| Gemini Edit | Quick style changes, creative edits | Fast, good for experimentation |
| Nano Banana 2 | General editing, background changes | Good balance of speed and quality |
The Best Workflow
The highest quality results usually come from combining image generation with 3D generation:
- Generate a reference image in Image Studio with a clean prompt
- Edit it if needed (remove background, adjust style, fix details)
- Convert to 3D using Image to 3D with Prism 3.1
This gives you more control than Text to 3D alone, because you can see and refine the 2D reference before committing to 3D generation.
Try Text to 3D - Open Image Studio - Try Image to 3D
Frequently Asked Questions
What’s the ideal prompt length for AI 3D generation?
For Text to 3D in 3D AI Studio, 5-20 words works best. Anything shorter than 5 words is usually too vague, and anything over 20 words tends to confuse the model. Focus on naming the object, its key material, and the style you want. A prompt like “medieval stone castle with tall tower” is more effective than a 50-word paragraph.
Do different AI models need different prompting styles?
Yes. In 3D AI Studio, Prism Turbo works best with short, punchy prompts (5-10 words), while Hunyuan 3.1 Pro handles more detailed descriptions. For image generation, Flux 2 prefers natural sentences over comma-separated keywords, and ImageGen 4 follows complex instructions well with minimal prompting. Check the model-specific tips table on this page for guidance.
How do I write prompts for game-ready 3D assets?
Add keywords like “game ready,” “game asset,” or “clean topology” to your prompt in 3D AI Studio. Specify the art style (“low-poly,” “stylized,” “realistic PBR”) and mention “one solid piece” to prevent floating geometry. For the Tripo P1 model specifically, just adding “game ready” to any object prompt optimizes the output for game engines.
Why does my AI 3D model have floating parts or broken geometry?
This usually happens when the prompt describes multiple objects or a complex scene. In 3D AI Studio, always describe a single object and add “one solid piece” or “single connected mesh” to your prompt. Avoid prompts like “a sword and shield” - generate each item separately instead.
What prompt keywords improve texture quality?
Adding “PBR texture,” “detailed surface,” or specific material names (“brushed steel,” “worn leather,” “polished marble”) in 3D AI Studio gives the AI clearer direction for textures. Also mention “realistic” or “photorealistic” if you want lifelike surface detail. Enable the PBR toggle in settings to get normal, roughness, and metallic maps alongside the base color.
How do I prompt for images that convert well to 3D?
Include “clean white background,” “single object, centered,” and “3D render style” in your image generation prompt. These keywords tell 3D AI Studio’s Image Studio to produce isolated objects with good depth cues that the Image to 3D converter handles easily. Avoid flat illustration styles, and use “full body” for characters to prevent cropping.