# The Ideogram 4.0 prompt guide Ideogram publishes a public search API over its indexed pages; the guide is at https://ideogram.ai/agents/agent-access. Source: https://ideogram.ai/blog/claude-mcp/ Source observed: 2026-11-08T15:45:23.127Z Last checked: 2026-11-08T15:45:23.127+00:00 Indexed: 2026-11-08T15:58:28.434+00:00 --- # The Ideogram 4.0 prompt guide **Published:** June 12, 2026 **Topic:** MCP and prompting Ideogram 4.0 with Claude We explain two prompting modes, style extraction from reference images, and our `/ideogram-prompt` Claude Code skill. [Ideogram MCP](https://ideogram.ai/features/mcp/) · [View the featured image on Ideogram](https://ideogram.ai/g/PSnsAZZcSYKuUjCIm28OMA/0) ## TL;DR - Install the Ideogram MCP in one command. It gives Claude `generate_image`, `describe_image`, `remix_image`, and collection management. - **Verbose prompting** lets our magic prompt layer interpret intent. - **Structured JSON prompting** bypasses that layer: the supplied structure is used as written. - `describe_image` returns a structured style recipe from a reference image, which Claude can apply to new subjects. - Save recipes as the `/ideogram-prompt` Claude Code skill for repeatable output across sessions. ## `/ideogram-prompt` skill The skill packages the techniques in this guide. Install it once and invoke it from any session. **Install command:** ```bash curl -fLo ~/.claude/skills/ideogram-prompt/SKILL.md --create-dirs https://ideogram.ai/skills/ideogram-prompt/SKILL.md ``` [Download the skill for desktop](https://ideogram.ai/skills/ideogram-prompt/SKILL.md). For the Claude desktop app, save the file to `~/.claude/skills/ideogram-prompt/SKILL.md`. ## Overview Ideogram 4.0 was trained on structured JSON captions. Plain-text prompts pass through a magic prompt layer that converts them to that format. JSON sent directly is used as written. The choice between these approaches is central to prompting Ideogram. The Ideogram MCP gives Claude direct access to `generate_image`, `describe_image`, `remix_image`, and collection management. Given a brief, Claude chooses a tool, structures the prompt, evaluates the output, and refines it. The workflow is a closed loop: describe, generate, evaluate, adjust, and try again. Claude retains context between steps. All images in the guide were generated in Claude Code with the Ideogram MCP. [Set up the MCP server](https://ideogram.ai/features/mcp/); the guide says setup takes about a minute. ## Verbose prompt vs. structured JSON ### Verbose — magic prompt on Give us a rich natural-language scene description. Our magic prompt layer interprets it and makes creative decisions about color, lighting, composition, and medium. This suits quick ideas, loose briefs, exploration, and cases where interpretation is welcome. ### Structured JSON — magic prompt off Use explicit `high_level_description`, `style_description`, and `compositional_deconstruction` fields for palette, medium, composition, lighting, and per-element descriptions. We use the JSON as written, giving direct control over palette, composition, text content, and element placement. This suits precise palettes, specific compositions, and legible text. ### Official v4 prompt schema The open-source [ideogram4 prompting guide](https://github.com/ideogram-oss/ideogram4/blob/main/docs/prompting.md) documents field ordering, bounding-box coordinates, color-palette conditioning, and worked examples. Key ordering matters because the model was trained on a specific sequence. - **Photo captions:** `aesthetics` → `lighting` → `photo` → `medium` - **Non-photo captions:** `aesthetics` → `lighting` → `medium` → `art_style` Use `color_palette` in `style_description` with up to **16 uppercase `#RRGGBB` hex values** to steer dominant colors, and up to **5 per element** for per-subject control. ### Wolf comparison The same subject is shown in both modes: - **Magic Prompt ON:** “a wolf in a snowy forest.” - **Magic Prompt OFF:** Claude-written v4 JSON with candid framing, a named amber-blue palette, authentic natural light, steam breath, and bokeh depth. ![Wolf in a snowy forest generated from a four-word brief with Magic Prompt on.](https://storage.googleapis.com/ideogram-static/website/images/094d630168040ff9.webp) ![Wolf in a snowy boreal forest, structured JSON prompt, candid editorial framing.](https://storage.googleapis.com/ideogram-static/website/images/3473b7b447bac51c.webp) Try this in Claude Code: ```text Generate a wolf in a snowy boreal forest. Write a structured v4 JSON prompt with magic prompt off — cool blue-grey and warm amber palette, cinematic editorial photograph framing, full body visible. Then show me the prompt you used. ``` ## Styles The style workflow is: describe a reference image, extract its visual logic, write a prompt recipe, and apply it to a new subject. Claude can save the recipe as reusable JSON. Each example below is sample output and a starting point; change the subject, palette, or composition. - **Blueprint + impressionist:** Blueprint linework on deep indigo vellum with Monet-style impressionist oil paint in the sky. [View on Ideogram](https://ideogram.ai/g/Podb6agvSpWQyBCEAZBOSA/0). Image: ![Haussmann apartment building elevation in precise blueprint linework on deep indigo vellum, with an impressionist oil-paint sky in warm violet and apricot breaking above the roofline.](https://storage.googleapis.com/ideogram-static/website/images/77187a4d83f5eaf4.webp) - **Memphis vanitas:** Dutch vanitas objects in oil glaze on a vivid Memphis surface. [View on Ideogram](https://ideogram.ai/g/LalhTYP8SgexqVCyKRBz-w/0). Image: ![Dutch Golden Age vanitas, skull, guttering candle, wilting rose, open Bible, painted in 17th-century oil technique, resting on a vivid Memphis design surface in yellow, magenta, and turquoise squiggles.](https://storage.googleapis.com/ideogram-static/website/images/69c7cc795da4cae0.webp) - **Constructivist + botanical:** Soviet constructivist poster geometry with botanical watercolor specimens. [View on Ideogram](https://ideogram.ai/g/DX9SUYimRcGYsdDIcwGjSw/0). Image: ![Soviet constructivist factory poster in deep red, black, and cream, with delicate botanical watercolor herbarium specimens, pansy, fern, lavender, mounted into the geometric composition.](https://storage.googleapis.com/ideogram-static/website/images/4c6702272d47be54.webp) - **Memphis still life:** Flemish Golden Age chiaroscuro still life on a Memphis tablecloth. [View on Ideogram](https://ideogram.ai/g/8U6hPzj_TVa51HqwLCAIdA/0). Image: ![Flemish 17th-century still life with oyster, lemon, olives, and pewter in full oil glaze chiaroscuro, arranged on a Memphis design tablecloth in coral, turquoise, and black squiggles.](https://storage.googleapis.com/ideogram-static/website/images/e72c33c3f79f04c7.webp) - **Topo map + portrait:** Photographic face and hands with a USGS topographic-map torso. [View on Ideogram](https://ideogram.ai/g/VYlbYNROTM6-0fCTlovc4g/0). Image: ![Environmental portrait of a weathered man whose torso becomes a USGS topographic map, contour lines following his anatomy, elevation numbers across the chest, while his face and hands remain photographic.](https://storage.googleapis.com/ideogram-static/website/images/d36329e84562f00c.webp) - **Screen print + 3D:** A photorealistic 3D object encoded with two-color Benday halftone dots. [View on Ideogram](https://ideogram.ai/g/VLnCJHevRhmqFeyyEnrQaQ/0). Image: ![Chess knight piece rendered as a photorealistic 3D object, its entire surface covered in two-color screen-print Benday halftone dots in deep teal and burnt orange, dot density encoding the volumetric depth.](https://storage.googleapis.com/ideogram-static/website/images/9b990311a268c6db.webp) ## Describe, extract, apply Claude can call `describe_image` on any image and receive structured JSON describing medium, palette, lighting, composition, grain type, and mark-making. Claude then synthesizes a style recipe and applies it to a new subject. The three sample reference images in this exercise share a vintage 1950s-era graphic sensibility: flat illustration, bold display type, limited spot colors, and a surreal or playful undercurrent. ### Extracted sample style recipe - **Medium:** Flat graphic illustration and bold display type; vintage event poster, botanical illustration, and retro typographic design; no 3D, photography, or gradients. - **Palette:** Three or four spot colors—forest green, butter yellow, aged cream, plus coral pink or black. - **Composition:** Dominant headline at top, central illustration, supporting detail below, with an element that breaks the formality. - **Texture:** Distressed print grain, risograph halftone, paper-stock tooth, and worn registration marks. Using that sample recipe, Claude generated three sample posters: - **THE GREAT ESCAPE. Horse + face-flowers.** ![THE GREAT ESCAPE vintage botanical event poster, galloping horse silhouette with face-flowers blooming from mane and tail, forest green and butter yellow.](https://storage.googleapis.com/ideogram-static/website/images/8b72335ce561029c.webp) - **WILD THINGS. Face-flower bouquet.** ![WILD THINGS vintage floral championship poster, face-flower bouquet illustration, aged cream background, forest green and coral pink.](https://storage.googleapis.com/ideogram-static/website/images/20307e44f38c3aab.webp) - **BLOOM OR WITHER. Type + face-flowers.** ![BLOOM OR WITHER typographic poster, flat bold condensed letters with face-flowers growing from terminals as living ornament, black ground, cream and green.](https://storage.googleapis.com/ideogram-static/website/images/96c31d436081770e.webp) The same process works from prose, such as: “1970s Italian horror movie poster, warm earth-tone palette, bold condensed display type, heavy grain.” Claude converts the description into JSON fields and generates without a reference image. ## Try it in Claude Code Connect the [Ideogram MCP](https://ideogram.ai/features/mcp/), open [Claude Code](https://claude.ai/code), and ask Claude to generate an image. Claude asks which mode to use. ## Gallery The gallery consists of sample images generated in one Claude Code session using the Ideogram MCP. The source provides descriptions and image URLs for a large collection, including botanical studies, ceramics, architecture, microscopy, caves, still lifes, book covers, textiles, typography, portraits, posters, natural-history plates, and screen prints. --- [Read this page’s summary](https://ideogram.ai/agents/api/public/v1/page-tldr?url=https%3A%2F%2Fideogram.ai%2Fblog%2Fclaude-mcp%2F&trace=t_380edc68d53d) ## Ask a follow-up about this page Need a more specific answer? Ask the exact question you have. It returns a fresh, cited answer from Ideogram’s knowledge base. No account or API key required. Open this URL with your own question, URL-encoded. Include your question in q. All research inputs are optional; no account or setup. `GET https://ideogram.ai/agents/agent/faq?q={question}&trace=t_380edc68d53d` [Tool guide and examples](https://ideogram.ai/agents/agent-access) · [OpenAPI](https://ideogram.ai/agents/api/public/v1/openapi.json) · [Page TLDR form](https://ideogram.ai/agents/agent-tools/page-tldr?url=https%3A%2F%2Fideogram.ai%2Fblog%2Fclaude-mcp%2F). Opening a form does not execute a tool.