
Code: @MiniMax_Code MiniMax Design: @Hailuo_AI API: https://xitter.twitterapi.xyz/t.co/fHRdSV73Hr M Plan: https://xitter.twitterapi.xyz/t.co/XfJWSyauda
San Francisco
Joined January 2025
- Tweets2.1K
- Following1K
- Followers131K
- Likes9.7K
Pinned Tweet
4K image generation in 10 seconds! ⚡
Meet MiniMax H3-Image.
16 million pixel resolution. Stunning realism. Precise prompt following & editing. Up to 16 reference images. All in 10 seconds.
And it's open-weight. 🚀
Available now via our API, MiniMax Code, and Hailuo AI — with 50% off 16MP+ generations during launch week!
Can't wait to see what you create. 🔥
Hugging Face: huggingface.co/MiniMaxAI/Min…
API: platform.minimax.io/docs/gui…
MiniMax Code: code.minimax.io/about
MiniMax (official) retweeted
After experiencing immersive film firsthand in August, I became determined to bring this concept to life.
Powered by MiniMax H3’s sota open-weight capabilities, fal & SGLang’s inference optimizations, and Pickford’s Emmy-winning creative expertise, we’ve made something special happen: Live Film - where audiences can direct the story and shape how it unfolds in real time.
We’re kicking off our global screening tour, with SGLang Summit as the first stop🥰Come experience the future of AI-powered interactive content firsthand!
Stay tuned guys!!!🥃
Replying to @sgl_project
When the audience changes the story.
Leanna Ren (@MiniMax_AI), Kenneth Loebenberg (@fal), Bernie Su (Pickford) and Mick Qian (@radixark) show how H3, post-training and diffusion inference power a live interactive mystery.
If your creative work moves fast, these two will keep up with you ⚡ @MiniMax_AI H3 Image and H3 Image Highspeed are live on Picsart, with 4K generation and editing in around 10 seconds and native ~16 MP output at any aspect ratio 🖼️
They also handle the details that usually slow things down, from multi-reference editing to long prompts and dense text for posters, infographics and UI screens 📝 Gaming, 3D and e-commerce creators, this one’s especially for you 🎮
Try here 👉 tinyurl.com/ye28pmcu
6
120
14,012
Start with a sketch. One product, every scene. Exactly as you prompt it.
Studio shot to lifestyle campaign, consistent in every frame. That's MiniMax H3 Image for e-commerce and ad creative at scale.
Thanks @eachlabs for showing it so beautifully ♥️ #H3Image
Native 4K images in about 10 seconds ⚡
MiniMax H3 Image, the new open-weight model by @MiniMax_AI, is now on each::labs.
It generates at up to 16MP in any aspect ratio, takes up to 16 reference images and edits with precision, from adding or swapping objects to rendering text, infographics and UI layouts.
Try it now on each::labs⤵
MiniMax (official) retweeted
100 days ago, we open-sourced MiniMax H3.
Now 1,472 models are built on it, downloaded 148M+ times on @huggingface
Thank you, builders of the H3 universe:
@ComfyUI @haoailab @fal @PrunaAI @video_rebirth @UtopaiStudios @HeyGen @ViggleAI @pika_labs @huggingface @ModelScope2022 @NVIDIAAI
huggingface.co/MiniMaxAI/Min…
🤖 Made with AI
Every image starts with a “what if.”
Here’s where we took it with MiniMax H3-Image. 🧵
What have you been making? Share your image & prompt in the replies 👇
Modern subjects in traditional artistic languages.
Prompt: Translate exterior maintenance on a modern space station into a Tang-dynasty Dunhuang-style earthen-wall mural in rich mineral pigments. The image is a flat wall painting, not a space movie. At the center, a technician resembling a flying apsara reaches sideways toward a solar-panel pivot, wearing a simplified modern safety vest over wide sleeves, with long ribbons forming two flowing arcs behind them. A companion at the lower left passes a tool to the main figure, while another at the upper right steadies a folded solar panel. All three direct their hands and eyes toward the same pivot. Clear, fine safety lines connect the figures to the station wall; ribbons and safety tethers each follow distinct paths and must not merge into tangled lines. The station consists of an ocher-white cylinder and a few geometric planes, with solar panels rendered as small azurite-blue squares. Make the modern facilities recognizable without smooth metal or realistic space lighting. Use Tang-mural-style linework for faces and clothing folds, light bodily contours, and overlapping forms instead of precise perspective. Render Earth at the side as a muted malachite-green disc. Keep the background earthen ocher with a few traditional scrolling clouds and star dots. Balance cinnabar red, azurite blue, malachite green, and warm white. Show mineral grains and slight natural wear in the paint layers, with locally exposed wall substrate rather than a uniform crackle filter. Organize movement through ribbons, panel wings, and the cylinder. Make the main figure large and the helpers slightly smaller, leaving open wall areas that breathe like blank paper. No sci-fi armor, blue volumetric light, photographic starfields, subtitles, trademarks, inscriptions, or seals.
9
2,474
Turning knowledge into visuals. H3-Image converts complex information into diagrams and posters.
Something harder? Turn a lengthy case narrative into a single evidence board, with timelines, positions, and testimonies grouped, annotated, and connected all in one frame.
Prompt: Create a detective case reasoning board to solve the following puzzle:
Confidential documents have been stolen from a safe. Exactly one of six suspects, A, B, C, D, E, and F, acted alone. Each makes one statement:
A: “The culprit is B or C.”
B: “D is not the culprit.”
C: “A's entire statement is false.”
D: “The culprit is E or F.”
E: “Exactly one of B's and D's statements is true, and the other is false.”
F: “A is not the culprit.”
Exactly three of the six statements are true and three are false. “Or” means at least one is true, and all references refer to the statements above.
Find the unique culprit and create a complete evidence board: show six letter-labeled suspect portraits and their exact statements, mark each statement true or false, and highlight the culprit. Include a concise verification table showing the number of true statements under each of the six possible culprit assumptions. Use connecting lines to show which statements refer to others, ensuring the portrait labels, reasoning, and table are consistent.
Do not add motives, alibis, accomplices, or extra evidence, and do not treat appearance or expression as clues. Use a cinematic film-noir detective style, with clear, in 16:9 landscape format.
1
1
861
Some moments should feel a little unplanned.
From creator.
Prompt: A cinematic, dynamic photograph with strong artistic expression, fusing slow-shutter panning photography, vintage sports documentary, and classic 35mm color film aesthetics.
The frame captures a real subject in high-speed motion — it could be a running animal, a cyclist, a moving car, a race vehicle, an athlete, or any other object with a clear directional trajectory.
The camera tracks the subject dynamically from the side or at a slight angle, using a slow shutter speed combined with a panning technique precisely matched to the direction of motion.
The subject's core silhouette remains recognizable to a degree, while body edges, rapidly moving limbs or components, and the surrounding environment take on a natural motion blur.
Motion blur is the dominant visual language of the image.
The background is stretched along the direction of movement into continuous, flowing horizontal bands of color — once-sharp trees, buildings, roads, and environmental details dissolve into abstract gradients of hue and light. The emphasis is on authentic optical motion trails, not a simple digital blur filter.
The overall composition retains distinct foreground, midground, and background layers, but allows certain regions to merge and overlap through motion, creating an effect closer to abstract painting.
The framing is restrained and cinematic: the subject sits near the center or slightly off-center, surrounded by generous negative space that makes the high-speed motion feel more solitary, elegant, and powerful.
Natural ambient light is used, preferably the low-angle sun of early morning or late afternoon, so the moving subject is lit by warm side light while the background holds deep, quiet shadows.
The color palette is a desaturated but tonally rich vintage film treatment, built around deep forest green, ink green, olive, caramel brown, golden yellow, and warm gray — the specific palette adapting naturally to the scene.
Emulates authentic 35mm color film texture: fine grain, soft sharpness, slight underexposure, naturally blooming highlights, and deep, muted shadow tones.
The image carries the sensibility of 1970s–1990s vintage sports photography and contemporary fine-art photo books — embracing chance, speed, real motion, and imperfect imagery. Rather than chasing the commercial, ultra-sharp frozen moment of an advertisement, it lets speed itself become the true subject of the frame.
Cinematic motion photography, artistic slow shutter panning, intentional motion blur, directional blur, horizontal light streaks, abstract background movement, vintage sports film photography, analog film grain, rich dark tones, warm natural highlights, imperfect focus, dynamic visual storytelling, fine art photography.
Avoid high-shutter frozen action, over-sharpening, artificial speed lines, exaggerated digital motion blur, cyberpunk neon, HDR, CGI rendering, and commercial sports-advertising polish.
Authentic photography, vertical 9:16 composition, no text, no watermarks.
A little dreamcore for your workday.
From creator.
Prompt: A surrealist photograph with intense poetic sensibility and quiet absurdity, fusing Dreamcore aesthetics, minimalist conceptual photography, vintage film texture, and contemporary fine-art photography.
In the middle of a boundless sea of white daisies, a middle-aged man in a light blue shirt and dark trousers sits alone on a modern office chair, diligently working at an old-fashioned CRT desktop computer.
In front of him sits a clean, deep blue-gray office desk, holding a heavy vintage computer monitor, a keyboard, and a few sparse office items.
His expression is calm and focused, as if entirely unaware of the strangeness of his surroundings. His posture is natural, with no exaggerated performance or forced humor.
The desk and chair are placed directly in the center of the flower field, with no surrounding office, road, or man-made structure — only an endless expanse of white daisies stretching to the horizon.
The composition is minimal and strongly symmetrical, shot from a distance. The figure and desk sit slightly below center, occupying only a small portion of the frame.
The flower field extends from the foreground all the way to the far horizon, where densely packed white daisies form a continuous, finely textured carpet.
The horizon line falls roughly one-third from the top of the frame, allowing the sky to occupy a large area of negative space — a soft, clean pale blue, free of complex cloud formations.
Warm late-afternoon sunlight enters from the side, casting soft golden highlights across the white petals, the man’s face, and the desk. The shadows carry a faint blue-violet hue, producing a warm-cool contrast that feels both tender and romantic.
The overall palette centers on sky blue, cream white, daisy yellow, soft grass green, and muted gray-blue — low contrast, slightly faded, soft and luminous.
Emulates Kodak Portra 400 color negative film: fine film grain, gentle highlight rolloff, natural skin tones, and slightly dreamlike color rendering.
Shot from a slightly distant observational perspective, with authentic natural light, realistic plant textures, and believable object shadows — lending the impossible scene the full credibility of documentary photography.
The overall mood is quiet, absurd, romantic, solitary, and tinged with dry humor — like a moment from a dream no one has bothered to explain.
Surreal everyday photography, dreamcore aesthetic, minimalist absurdism, conceptual fine art photography, lonely office worker in a field of daisies, vintage CRT computer, endless flower meadow, pastel blue sky, warm afternoon sunlight, subtle analog film grain, poetic surrealism, cinematic realism.
Vertical 9:16 composition, no text, no watermarks.
5
680
MiniMax H3 × @HeyGen 🎬
Let your AI agent direct. Make your best cinematic shot through the HeyGen MCP with HeyGen Video 1, which is built on MiniMax H3, and win $5K in MiniMax API credits.
Post with #HeyGenVideoContest by Dec 1 👇
heygen.com/blog/video-contes…
We're teaming up with @Runware, @MiniMax_AI, @ComfyUI and @ChromaAwards to give away $20K+ in cash and credits for the best videos made with HeyGen Video 1 through our MCP.
To enter, follow @HeyGenMCP and post your video with #HeyGenVideoContest.
Rules: heygen.com/blog/video-contes…
Huge thanks to @runware for Day 0 support of MiniMax H3 Image 💚
MiniMax H3 Image: 16 references in, 2–3x faster out. Built for ad creative, e-commerce, and content at scale.
#H3Image
Replying to @MiniMax_AI
@MiniMax_AI H3 Image & H3 Image Highspeed are live on Runware on D0
MiniMax built its new image model on H3, its popular open-weight video model, calling it the strongest open-weight image model released so far.
- Edit with up to 16 reference images
- Output up to 20 MP, with 1K, 2K and 4K presets or exact width and height
- Aspect ratios from 1:3 to 3:1
- H3 Image Highspeed runs at 2x the speed, for quick edits and fast iterations
Image output from $3.80 per million tokens, with 50% off 4K outputs on both models through November 17
MiniMax H3 Image × @fal 🚀
4K images in ~10 seconds. Follows complex prompts, edits with precision, and handles game and 3D assets with ease. Fast enough to power agents and productivity tools, not just image apps.
Thanks fal team, always a pleasure building with you.
#H3Image
MiniMax H3 Image is now available on fal.
The latest open-weights image model from @MiniMax_AI delivers exceptional visual quality at lightning-fast speeds, with powerful generation and editing capabilities.
Generate and edit images in native 4K, combine multiple references, and render complex text with precision.
We're teaming up with @Runware, @MiniMax_AI, @ComfyUI and @ChromaAwards to give away $20K+ in cash and credits for the best videos made with HeyGen Video 1 through our MCP.
To enter, follow @HeyGenMCP and post your video with #HeyGenVideoContest.
Rules: heygen.com/blog/video-contes…
M3.1 Flash is live on @FireworksAI_HQ . Let’s talk about how it was built—over cocktails.
Join MiniMax and @FireworksAI_HQ for Code & Cocktails in SF, where @Prof_OZ from Fireworks and @VictorSuOrtiz from MiniMax will discuss:
→ What went into building M3.1 Flash for long-running agents
→ How coding, tool use, and visual reasoning work together
→ What cheaper cached tokens mean for the cost of agent loops
Then drinks, bites, and time to compare notes with other builders.
📅 Tuesday, November 10 · Doors at 6 PM, fireside at 6:45 PM
📍 San Francisco · Venue shared after registration approval
Bring your questions. Registration below.
#MiniMax #FireworksAI #OpenModels #AgenticAI #AIEngineering
MiniMax M3.1 Flash, on more of your hardware.
Quantizing a 428B model opens up more ways to run it locally. Lower-bit weights reduce the memory needed to load the model and the weight data moved during decoding. Making those savings useful takes careful calibration, efficient low-bit kernels, and hardware-aware placement of experts across GPU and system RAM.
Thank you to the builders doing that work:
• @UnslothAI @danielhanchen — 2-bit Dynamic GGUF weights at 143.9 GB, with calibration and fidelity testing across coding, tool use and other workloads.
• @ViC305 — SAGE-EXL3, token-level fidelity measurements and a single-DGX-Spark recipe.
• @PeasantSmith — Project Maya quants and expert offloading for local coding and tool use.
• @MiaAI_lab — EXL3-EAQ quantization work.
• @jundotkim + mlx-community — oQ quants and DSpark-powered decoding through oMLX on Apple Silicon.
• @AIatAMD — Quark MXFP4 adapted for AMD Instinct.
• @atomic_chat_hq — GGUFs for local desktop workflows.
• @superalesha — custom inference work combining quantized weights, expert caching and system RAM on 4× RTX 3090.
Every shared quant, kernel, measurement and recipe gives the next builder a better starting point. Thank you.
















