Chrome Extension
Ideogram just changed what open-weight image generation means. On June 3, 2026, they released Ideogram 4.0, their first open-weight text-to-image model, and the numbers it put up immediately made me sit down and rethink my entire image generation workflow.
I have been testing every prompting pattern in the model, working through the JSON prompting system, pushing the bounding-box controls, and running the typography capabilities against real client-work standards. This guide is the complete breakdown: the architecture, the benchmarks, the API pricing, every prompting trick that changes the output quality, and 22 ready-to-use prompts organized by use case.
If you generate images professionally and you have not tested Ideogram 4.0 yet, this article covers everything you need to start today.
Checkout our Updated and free Ideogram Prompt generator here.
Key Takeaways
What Is Ideogram 4.0? Architecture and Technical Specs
Ideogram 4.0 is a foundation model built entirely from scratch by Ideogram AI, a research lab focused on image generation with a particular strength in typography and design output. It is not a fine-tune or distillation of any existing model. Every parameter was trained from the ground up.
| Specification | Detail |
|---|---|
| Release date | June 3, 2026 |
| Parameters | 9.3 billion |
| Architecture | Fully single-stream Diffusion Transformer (DiT), 34 layers |
| Text encoder | Qwen3-VL-8B-Instruct (vision-language model) |
| Encoder layers used | Hidden states from 13 intermediate layers, concatenated |
| Native resolution | 2048 x 2048 (2K) |
| Resolution range | 256 to 2048 (multiples of 16) |
| Max aspect ratio | 6:1 |
| Guidance type | Dual-branch classifier-free guidance |
| Prompt format | Structured JSON (native), plain text via Magic Prompt |
| License | Non-commercial (open-weight) |
| Quantizations available | nf4 (CUDA), fp8 (all hardware) |
The architectural choice that matters most is the text encoder. Instead of CLIP or T5, the model uses Qwen3-VL-8B-Instruct, a full vision-language model. Hidden states are extracted from 13 of its intermediate layers and concatenated, giving the model multi-scale semantic features that range from surface-level token understanding to deep compositional reasoning. This is why Ideogram 4.0 handles complex, densely described prompts better than models using traditional text-only encoders.
The fully single-stream DiT architecture means text and image tokens are concatenated into one unified sequence and processed through the same 34-layer transformer.
There are no separate text or image branches. Every transformer layer sees both modalities simultaneously, which enables the model to develop genuinely cross-modal representations rather than aligning two separate streams at the output stage.
Benchmark Performance: What the Numbers Show
Ideogram 4.0 has been evaluated across third-party arenas, open-source benchmarks, and an internal professional design evaluation. The results are consistent across all of them.
Design Arena
On Design Arena, a third-party image ELO leaderboard specifically focused on design-quality generation, Ideogram 4.0 is the top-ranked open-weight model overall, trailing only GPT-Image-2 and Gemini models from companies with vastly larger compute budgets. Among open-weight models only, it leads by a commanding margin ahead of the next-best open model.
ContraLabs Typography Evaluation
ContraLabs ran a blind typography evaluation judged by ten professional designers from Contra's top-earning talent pool. Ideogram 4.0 dominated:
| Model | First-Place Win Rate | Real Client Work Score |
|---|---|---|
| Ideogram 4.0 | 47.9% | 3.55 / 5 |
| Nano Banana 2 (Gemini) | 30.0% | 2.84 / 5 |
| FLUX.2 max | 15.5% | 2.49 / 5 |
| Grok Imagine 1.0 | 15.0% | 2.61 / 5 |
The real client-work score is the number I trust most from this evaluation. It measures a more practical question than win rate: would a professional actually use this output for a paying client? Ideogram 4.0 scored 3.55, a gap that reflects genuine production quality, not just benchmark gaming.
Open-Source Benchmarks
On standard benchmarks covering layout control (7Bench), spatial reasoning and object fidelity (SpatialGenEval), text rendering (X-Omni OCR), and prompt alignment (Prism):
LMArena and Internal Evaluation
On LMArena, Ideogram is the top-ranked open-weight lab and a top-5 image generation lab overall. In Ideogram's own internal human-preference benchmark, judged blind by professional graphic designers, Ideogram 4.0 ranks number two overall behind only GPT-Image-2 medium, and is the top open-weight model.
Pricing: API and Subscription Tiers
Web Platform Subscriptions
| Plan | Monthly Cost (Annual) | Priority Generations | Private |
|---|---|---|---|
| Basic | $8/month | 400/month | 100/month |
| Plus | $20/month | 1,000/month | Unlimited |
| Pro | $60/month | 3,000/month | Unlimited |
All plans include unlimited slow generations in addition to the priority quota. The Pro plan also includes API discounts.
API Pricing
The API is available at developer.ideogram.ai. API access requires setting up an account at ideogram.ai/manage-api, accepting the Developer API Agreement, adding payment information, and creating an API key.
| Quality Tier | Price Per Image |
|---|---|
| Turbo | $0.03 |
| Standard | ~$0.06 |
| Quality | $0.09 |
Volume-based discounts are available for annual commitments. For high-volume production workloads, contact Ideogram directly at ideogram.ai/features/api-pricing.
The API supports image generation, editing, upscaling, background removal, and remixing. The Turbo tier is suitable for high-volume workflows where speed matters. The Quality tier is for final production assets where fidelity is the priority.
The JSON Prompting System: The Core Differentiator
Most image models accept plain-text prompts. Ideogram 4.0 was trained on structured JSON captions. This is not a convenience layer added on top of a conventional model. The JSON prompting system is native to how the model was trained, and it is what gives the model its precision.
Training captions exhaustively describe every element in the image. The more relationships a caption pins down, the more grounded supervision the model extracts from a single training pair. That is why the JSON prompt structure produces qualitatively different outputs compared to plain-text prompts on the same model.
The Three JSON Controls Worth Knowing
Bounding-box layout. Any element can be placed by specifying its bounding box as [y_min, x_min, y_max, x_max] in 0-1000 normalized coordinates. This means you can specify that a logo appears in the upper-left quadrant, a headline appears in the lower center, and a product image fills the right half. The model places elements where you tell it, not where it decides to put them.
Color palette conditioning. You can specify a colour_palette array of up to 16 hex colors per image, and up to 5 hex colors per individual element. The model uses these to steer the dominant color scheme of the output. Combined with bounding boxes, this gives you compositional and chromatic control from a single prompt.
Typed text elements. Text elements in the JSON carry two pieces of information: the literal string to render, and a separate visual description for styling. This enables multi-line, multi-font in-image text with distinct styling per element. A headline can use a different font treatment than a subheadline or a caption, all specified in one prompt.
Magic Prompt: JSON Without Writing JSON
If you do not want to write JSON by hand, Magic Prompt handles the conversion. It is a hosted LLM expansion layer that takes your plain-text description and rewrites it into a full structured JSON caption before generation. It is free to use via Ideogram's hosted API (requires an API key from ideogram.ai). The expansion runs server-side, so no local model is needed.
Magic Prompt is the recommended starting point for casual use. For production workflows where you need precise control over layout, color, and typography, write the JSON directly.
Key Prompting Tricks That Change Outputs
Trick 1: Keep Plain-Text Prompts Under 80 Words
When using plain-text prompts (without JSON), the model processes the tail end of long prompts less reliably. Prompts over 90 words risk having later-listed elements dropped entirely. Trim to under 80 words, and put your most important elements first.
Trick 2: Replace Vague Adjectives with Visual Specifics
The model does not have a reliable visual reference for descriptors like "beautiful," "interesting," or "modern." These terms generate inconsistent outputs because they are not anchored to observable visual properties. Replace them with specific material, lighting, and composition descriptions.
Vague: "a beautiful product photo of a bottle" Specific: "a 100ml amber glass dropper bottle on a white marble surface, soft diffused studio lighting, casting a faint shadow to the right, centered composition"
Trick 3: Name the Art Movement or Style Reference
Using named art styles, movements, or photographic techniques gives the model a rich cluster of visual associations to work from. "Bauhaus poster design," "brutalist typography," "golden-hour editorial photography," "risograph print aesthetic," and "isometric flat illustration" all produce significantly more consistent outputs than broad terms like "artistic" or "graphic."
Trick 4: Toggle Magic Prompt Off When Precision Matters
Magic Prompt is useful for exploration, but it rewrites your prompt before generation. If you have a critical detail that must appear, such as a specific piece of text, a precise layout, or a brand color, Magic Prompt may rewrite it in a way that loses that detail. Turn Magic Prompt off and run the precise prompt directly.
Trick 5: Use the Sampler Preset for Quality
When running locally, set --sampler-preset V4_QUALITY_48 and --height 2048 --width 2048 for the highest-quality output. The default sampler is faster but does not produce the same fidelity as the quality preset at full 2K resolution.
Trick 6: Use Bounding Boxes for Text-Heavy Layouts
For any output where text placement matters, such as posters, packaging, business cards, and social media ads, use bounding-box coordinates to anchor text elements. Without explicit placement, the model places text in positions that are compositionally reasonable but may not match your specific layout requirements.
Trick 7: Specify the Negative Space Intentionally
Instead of using a negative prompt to exclude unwanted elements, describe the negative space you want as positive instruction. "Clean white background, no other objects in frame" is more reliable than a negative prompt listing everything you do not want. The model was trained on descriptive captions, not exclusion lists.
Trick 8: Describe Lighting as a Source
Generic lighting terms like "well lit" produce unreliable results. Lighting instructions that specify the source, direction, and quality produce consistent outputs: "single soft key light from the upper left with a subtle fill from the right," "natural window light falling from above right," or "warm practical light from a candle at frame center." Treat the model like a photographer: tell it where the light comes from.
22 Use Cases with Prompts
Poster and Print Design
Use case 1: Concert poster with hierarchy
{
"global": {
"description": "Concert poster, dark matte black background with textured grain",
"colour_palette": ["#0A0A0A", "#C9A84C", "#FFFFFF", "#E8D5A3"]
},
"elements": [
{
"description": "Large headline text in art deco serif",
"text": "VELVET UNDERGROUND REVIVAL",
"bbox": [50, 80, 200, 920]
},
{
"description": "Atmospheric illustration of a microphone stand, golden spotlight from above",
"bbox": [200, 150, 700, 850]
},
{
"description": "Date and venue text, small weight, centered",
"text": "August 14, 2026 | The Fillmore, San Francisco",
"bbox": [720, 100, 800, 900]
},
{
"description": "Website in small sans-serif",
"text": "velvetrevival.com",
"bbox": [900, 350, 950, 650]
}
]
}Use case 2: Motivational quote poster
A typographic poster on a deep forest green linen-texture background. Centered large serif display font in warm cream: "DO THE WORK THAT SCARES YOU." Below in a thinner weight, smaller size: "Because that is where growth lives." A single thin horizontal rule between the two lines in gold. No illustrations. Minimalist fine art print aesthetic.
Logo and Brand Identity
Use case 3: Wordmark logo concept
A wordmark logo for a law firm called "HARLOW & COLE". Deep navy blue (#1B2A4A) background. The firm name in a sharp, authoritative serif typeface in off-white (#F5F0E8). Below the name in smaller spaced uppercase sans-serif: "ATTORNEYS AT LAW". A single thin horizontal rule in muted gold between the name and descriptor. Clean white padding around all elements. Professional, authoritative brand identity.
Use case 4: Icon and wordmark lockup
Brand identity lockup on white background. Left side: a geometric icon consisting of two overlapping triangles forming a diamond shape in electric blue (#2563EB). Right side: wordmark "NEXFIELD" in bold black sans-serif, and below it "Data Infrastructure" in lighter weight grey. Ample white space between icon and text. Modern B2B technology brand aesthetic.
Social Media Assets
Use case 5: Instagram story with CTA
Instagram story, 9:16 ratio. Gradient background from deep violet at top to midnight blue at bottom. Large bold centered white headline: "LIMITED SPOTS LEFT". Below in medium weight: "Join 500+ founders building with AI". A bright coral CTA button at the bottom third: white text "APPLY NOW" inside. Small URL text below button: "foundry.ai/apply". No images, typography-only design.
Use case 6: Twitter/X card header image
Wide horizontal Twitter header image, 3:1 ratio. Textured dark charcoal background. Left-aligned bold white headline: "I write about building startups with AI tools." Below in lighter weight in light grey: "3x founder | 12,000+ subscribers | Every Tuesday." A subtle geometric pattern of small dots in very dark grey fills the right half of the background. Clean, personal brand aesthetic.
Use case 7: LinkedIn banner
LinkedIn profile banner, wide horizontal. Deep teal gradient background (#0D4F4F to #1A7A7A). Right-aligned white bold headline: "Product Designer". Subtext below in lighter weight: "UX | Systems | Strategy". Left side: abstract minimal illustration of overlapping circles and rectangles in translucent white. Professional, creative industry aesthetic.
Product Packaging
Use case 8: Coffee bag front label
Front label for a specialty coffee bag. Kraft paper texture background. Centered bold vintage-style serif in dark brown: "SUMMIT ROAST". Below in smaller caps: "SINGLE ORIGIN ETHIOPIA YIRGACHEFFE". Center illustration: a detailed hand-drawn engraving style mountain peak with coffee plants in the foreground. Roast level strip at the bottom: "MEDIUM ROAST" in small white text on a dark stripe. Artisan packaging aesthetic. Warm brown, cream, and terracotta color palette.
Use case 9: Cosmetics product box
Luxury skincare box flat design, front panel. Soft sage green (#A8C5A0) matte background. Centered gold foil-effect italic serif text: "LUMINAE". Below in small elegant serif: "Vitamin C Brightening Serum". A minimal botanical line illustration of a citrus branch in pale gold below the product name. Net weight in small text: "30ml / 1 fl oz". Luxury cosmetics packaging, understated and premium.
Business Cards
Use case 10: Minimal professional card
Business card flat design, horizontal orientation, 3.5 x 2 inch. Matte black background. White bold text left-aligned: "SARAH CHEN" on the first line, "Creative Director" lighter weight below. Right side in small grey text: "sarah@studioform.io" and "+1 628 555 0142". A single thin white horizontal rule across the lower quarter. Ultra-minimal, premium brand identity aesthetic. No logo, typography only.
Use case 11: Bold two-color card
Business card, square format, 2.5 x 2.5 inch. Split design: left half solid black, right half solid electric yellow (#FFE500). On the black half: white bold uppercase sans-serif "JAMES". On the yellow half: black bold uppercase "OKORO". Below the split on the full card: small black text on white strip "Brand Strategist | jamesokoro.co". High contrast, editorial design aesthetic.
Poster Typography Art
Use case 12: Typographic art print
Typographic art print, portrait format. Flat terracotta background (#C8674A). The word "BREATHE" in an ultra-wide, heavy-weight display typeface spanning almost the full width of the composition. The letters are hollow outline-only in white. Inside each letterform: fine botanical illustration in white line work, different plant for each letter. Minimal signature in small text at the bottom right: "Type + Nature Vol.3". Museum-quality art print aesthetic.
Book and Publication Design
Use case 13: Non-fiction book cover
Non-fiction book cover, portrait orientation. Dark navy background (#0D1B2A). Bold white headline title centered, two lines: "THE LEVERAGE" on the first line, "POINT" on the second in larger size. Below the title in smaller weight grey: "How Decisions Compound Into Outcomes". A central geometric illustration of interlocking lever shapes in gold. Author name at the bottom: "MICHAEL TRAN" in clean sans-serif. Publisher imprint at the very bottom in very small text. Award-winning non-fiction cover aesthetic.
Use case 14: Magazine feature spread hero
Wide landscape magazine article hero image. An aerial photograph-style view of a glass skyscraper at dawn, the glass facade reflecting the orange-to-pink sunrise sky. Overlaid bold white headline text in the lower third: "THE ARCHITECTURE OF TOMORROW". Below in lighter weight: "How AI is redesigning the way we build cities." Subtle dark vignette on the lower portion to make white text legible. Architectural magazine editorial aesthetic.
T-Shirt and Merchandise Design
Use case 15: Vintage band-style tee graphic
T-shirt graphic design on white background, ready for print. Distressed vintage aesthetic, printed look with age texture. Center graphic: a detailed illustration of an eagle with wings spread, lightning bolts in its talons, in dark navy blue ink. Above the eagle in curved vintage type: "PACIFIC NORTHWEST". Below in straight type: "EST. 1971". Underneath: a small mountain range silhouette. Classic Americana graphic tee aesthetic. Print-ready illustration.
Use case 16: Minimalist typography hoodie graphic
Apparel graphic design, screen print style on transparent background. Center-chest placement. Three lines of bold condensed sans-serif type: "BUILD" on the first line in white, "SHIP" on the second in bright yellow, "REPEAT" on the third in white. Each word in all-caps, slightly different sizes creating rhythm. No illustration. Clean, modern streetwear aesthetic.
UI and App Mockups
Use case 17: App store screenshot banner
Mobile app store screenshot banner, 6.5-inch format. Dark mode UI background. Top half: phone frame showing a clean dashboard UI with a circular progress ring in blue and a goal tracker. Below the phone: bold white headline "Hit Every Goal." Subtext in light grey: "Track. Focus. Achieve." Blue gradient accent behind the phone frame. Consumer app store marketing aesthetic.
Use case 18: SaaS landing page hero mockup
Wide desktop landing page hero section mockup. Dark background (#0F172A). Left side: large white bold headline across two lines: "Your Data," then "Your Control." Below in light grey: "Enterprise analytics without the enterprise complexity." Two CTA buttons: solid blue "Start Free Trial" and ghost white "See Demo". Right side: a screenshot of a clean dark-mode dashboard interface with bar charts and metric cards. B2B SaaS landing page aesthetic.
Photography-Style Output
Use case 19: Lifestyle product portrait
A lifestyle photograph-style image of a man in his early 30s sitting at a wooden cafe table. He holds a small espresso cup. Wearing a cream linen shirt, slightly rolled sleeves. Soft morning window light from the left. A laptop and notebook visible but out of focus in the background. Expression: relaxed, slightly smiling. Shallow depth of field, warm color grade. Editorial lifestyle photography aesthetic.
Use case 20: Architecture exterior visualization
Architectural visualization of a contemporary house, exterior view. Single-story volume, white rendered walls, floor-to-ceiling windows, flat roof with a narrow overhang. Set in a landscaped garden with ornamental grasses and a single mature olive tree to the left. Photographed at golden hour, warm light casting long shadows. Slightly elevated three-quarter angle. High-end architectural photography aesthetic. Photorealistic rendering.
Pattern and Texture Design
Use case 21: Seamless surface pattern
A seamless repeating surface pattern design. Tropical botanical print: large monstera leaves in deep green, birds of paradise flowers in orange and purple, small white star-shaped flowers as fill. Rich, dense composition with no visible background. Style: mid-century modern tropical illustration. Suitable for fabric or wallpaper. Square tile, pattern edges match.
Infographic and Data Visualization
Use case 22: Process infographic
A clean horizontal process infographic on white background. Four steps in a left-to-right flow connected by arrows. Each step: a circular icon in soft blue (#3B82F6) above bold black step number, below that a short label in bold and a two-line description in light grey. Step 1: "RESEARCH" icon is a magnifying glass. Step 2: "DESIGN" icon is a pencil. Step 3: "BUILD" icon is a code bracket. Step 4: "LAUNCH" icon is a rocket. Flat vector infographic style. No shadows, no gradients except icon circles.
UGC Characters
Use case 23: Realistic Selfie Shots
Use case 24: Realistic Mirror selfie
Graphs and Infographics
Use case 25: interactive Charts
How to Access Ideogram 4.0
Web platform. The fastest way to start is at ideogram.ai. All subscription plans include access to Ideogram 4.0 as the default model. The Basic plan at $8 per month on the annual billing is the entry point, with 400 priority generations per month.
API access. Set up developer access at ideogram.ai/manage-api. Accept the Developer API Agreement, add payment, and create your API key. Full API documentation at developer.ideogram.ai. Pricing starts at $0.03 per image on the Turbo tier.
Self-hosted via Hugging Face. The model weights are gated on Hugging Face. Access the nf4 version at huggingface.co/ideogram-ai/ideogram-4-nf4 or the fp8 version at huggingface.co/ideogram-ai/ideogram-4-fp8. Accept the license gate, authenticate with an HF token, and run inference with the provided CLI.
GitHub / local inference. The full inference codebase is at github.com/ideogram-oss/ideogram4. Install with pip install ., run run_inference.py with your prompt and API key for Magic Prompt expansion.
ComfyUI. Ideogram 4.0 has day-0 support in ComfyUI for developers who use node-based diffusion workflows.
Ideogram 4.0 vs the Competition
| Model | Parameters | Open Weight | Text Rendering | Layout Control | API Price/Image |
|---|---|---|---|---|---|
| Ideogram 4.0 | 9.3B | Yes (non-commercial) | Best open-weight | Bounding box native | $0.03-$0.09 |
| FLUX.2 dev | 32B | Yes | Below Ideogram 4 | Limited | Varies |
| HunyuanImage 3.0 | 80B MoE | Yes | Below Ideogram 4 | Limited | Varies |
| Qwen-Image | 20B | Yes | Below Ideogram 4 | Limited | Varies |
| GPT-Image-2 | Closed | No | Top overall | No bounding box | ~$0.04-$0.08 |
| MAI-Image-2.5 | Closed | No | Strong | No bounding box | $0.047 |
The comparison that matters most: Ideogram 4.0 at 9.3B parameters produces better text rendering than FLUX.2 dev at 32B, HunyuanImage 3.0 at 80B MoE, and Qwen-Image at 20B. The architectural choice of a vision-language text encoder (Qwen3-VL-8B) and JSON prompting training is doing more per parameter than conventional approaches.
The one area where GPT-Image-2 still holds an edge is on photorealistic human portraiture. For design work, typography, posters, packaging, branding, and commercial product imagery, Ideogram 4.0 is the model I reach for.
Frequently Asked Questions (FAQs)
What is Ideogram 4.0 and is it free to use?
Ideogram 4.0 is an open-weight 9.3B parameter text-to-image model released by Ideogram AI on June 3, 2026. The web platform at ideogram.ai offers a free tier with limited slow generations per month. Paid plans start at $8 per month on the annual billing for 400 priority generations. The open weights on Hugging Face are free to use under a non-commercial license.
What makes the JSON prompting system different?
Ideogram 4.0 was trained exclusively on structured JSON captions, not plain text. The JSON format allows you to specify bounding-box coordinates for precise element placement, hex color palette conditioning for exact color control, and separate text-string and visual-styling descriptions for each text element in the image. This gives you layout-level control that plain-text prompts on other models cannot replicate.
How does Ideogram 4.0 perform on typography compared to other models?
In the ContraLabs blind evaluation by professional designers, Ideogram 4.0 was picked as the best output 47.9% of the time on typography, well ahead of Nano Banana 2 at 30.0%, FLUX.2 max at 15.5%, and Grok Imagine 1.0 at 15.0%. On the real client-work usability score, designers rated it 3.55 out of 5 versus 2.84 for the nearest competitor. On standard benchmarks, it has the best text rendering of any open-weight model at any parameter count tested.
Can I run Ideogram 4.0 locally?
Yes. The weights are available on Hugging Face in nf4 and fp8 quantizations. The nf4 version runs on CUDA hardware with Diffusers support. The fp8 version runs on all hardware without Diffusers. You need a Hugging Face account, accepted license gate, and an authenticated HF token. The inference code and documentation are at the GitHub repository.
What is the difference between the Turbo and Quality API tiers?
The Turbo tier at $0.03 per image is optimized for generation speed and is suitable for high-volume workflows where you prioritize throughput. The Quality tier at $0.09 per image is optimized for maximum fidelity and is the right choice for final production assets. Both tiers use the same underlying Ideogram 4.0 model. The difference is in the inference configuration, including sampler steps and guidance scale.
What aspect ratios does Ideogram 4.0 support?
The model supports any resolution from 256 to 2048 pixels on each side, in multiples of 16, with aspect ratios up to 6:1. This covers square images, portrait and landscape orientations, ultrawide banners, phone wallpapers, and social media formats in a single model. The noise schedule auto-adjusts per resolution.
Final Thoughts
Ideogram 4.0 is the open-weight model I have been waiting to arrive. The combination of a 9.3B parameter model that beats every larger open-weight competitor on text rendering and layout control, with an accessible pricing structure starting at $8 per month and $0.03 per API image, makes it practical for production workflows that previously required proprietary models.
The JSON prompting system takes twenty minutes to understand and then becomes the feature you will use on every serious output. The bounding-box controls alone are worth the learning curve for anyone producing posters, packaging, business cards, or any asset where element placement matters.
Start at ideogram.ai to test the web platform with the prompts in this guide. Move to the API when you are ready to automate. If you want to self-host or fine-tune, the weights and code are at GitHub and Hugging Face.
The technical blog post at ideogram.ai/blog/ideogram-4.0 has the full architecture breakdown if you want to go deeper on how the DiT and vision-language encoder interact.
