P-H3-Image is Pruna AI’s open-weights image model based on MiniMax H3. It generates text-to-image stills and performs instruction edits with up to eight input images.

Is P-H3-Image the same as MiniMax H3 Image?

No. MiniMax H3 Image is MiniMax’s own image generation and editing surface. P-H3-Image is a separate open model derived from MiniMax H3 technology, with different reference limits and sizing guidance.

What resolution works best?

Text-to-image is strongest around 8 megapixels. Edits often look best at 4 or 8 MP. Lower resolutions can introduce more artifacts in faces, hands, and fine detail.

How do multi-image edits work?

Picture 1 is the image being edited. Additional images become Picture 2, Picture 3, and so on, so your instruction can borrow a hat, background, or prop from a named reference.

What is the prompt upsampler?

By default, short requests are expanded into fuller prompts before rendering. That helps conversational edits, but you can bypass it when you already have a complete prompt.

What are the main limitations?

Hype-heavy or awkward prompt lengths can weaken results. Hands and tiny background figures may distort, especially below 8 MP. Dense text is unreliable; short quoted strings work better. P-H3-Image does not generate video or audio.

P-H3-Image AI Image Generator

Explore P-H3-Image for photoreal, editorial, and cinematic stills. Built by Pruna on MiniMax H3 technology, it targets efficient 8-step generation around 8 megapixels, with instruction edits that can pull from up to eight input images.

8
input images
8 MP
sweet spot
8
sampling steps
T2I
+ instruction edit

Explore P-H3-Image AI Image Generation

P-H3-Image is Pruna AI’s open image model derived from MiniMax H3. It is designed for creators who want photoreal and film-like stills without treating every prompt like a giant video stack—efficient sampling, a built-in prompt upsampler, and instruction edits that can compose multiple pictures into one frame.

Photographers, editorial teams, and indie makers use P-H3-Image to draft portraits with real skin texture, documentary looks, graphic frames with short quoted text, and multi-image composites from a one-line request.

Explore P-H3-Image

What Is P-H3-Image?

P-H3-Image is an open-weights image model by Pruna AI based on MiniMax H3. It covers text-to-image and instruction-based editing in one checkpoint, with sampling tuned for roughly eight steps and no classifier-free guidance.

It shines at photorealism, film and editorial looks, and graphic styles with short quoted text. A prompt upsampler is included by default so short requests become the fuller prompts the model prefers.

P-H3-Image is not MiniMax’s hosted H3 Image product. Treat it as a related open model in the H3 lineage—useful for efficient high-megapixel stills and multi-image composites, with different reference limits and sizing habits.

8 MP sweet spot

Text-to-image looks strongest around eight megapixels; lower sizes show more artifacts.

Prompt upsampler included

One-line requests expand into the detailed briefs P-H3-Image renders best.

Edits with up to 8 images

Picture 1 is edited; Picture 2+ are references you can name in the instruction.

Photorealism, Film Stills, and Short In-Image Text

P-H3-Image is strongest where texture and lighting have to feel observed rather than plastic: documentary photography, editorial portraits, anime frames, drawings, prints, and graphic design with short quoted strings.

“1970s film still of a quiet coastal highway at dusk, warm sodium lamps, light film grain, medium telephoto compression, natural skin texture if people appear, no logos, no dense body copy.”

The built-in prompt upsampler turns a short request into the fuller brief the model renders best. That helps when you want to stay conversational—“make the pickup truck red”—without hand-writing every lighting note.

Keep expectations honest on dense typography: short quoted text works better than body copy or code. Hands and tiny background figures improve as you move toward the 8 MP sweet spot.

Why 8 Megapixels Matters for P-H3-Image

Text-to-image is tuned around 8 MP canvases; lower resolutions can work, but artifacts show up more often in faces, hands, and fine detail.

Common 8 MP frames include square ~2912×2912, 16:9 ~3872×2176, and 9:16 ~2176×3872. Sides are typically rounded to multiples of 32 in open pipelines.

Edits usually land best at 4 or 8 MP. Without an explicit canvas, many pipelines generate at the size of Picture 1, scaled up toward a 4 MP floor when the source is smaller.

Instruction Edits with Up to Eight Images

Pass one image to edit it, or a list of up to eight. Picture 1 is the image to change; Picture 2, Picture 3, and later frames act as named references.

A favorite community pattern is composing several portraits or props into one scene from a single sentence—the upsampler names what to take from each picture, and P-H3-Image rebuilds the frame.

“Give the woman in picture 1 the hat from picture 2, keep her pose and face, soft window light, photoreal, no text.”

State what must stay fixed. Without that guardrail, instruction edits can redraw more of the scene than you intended.

Where Creators Use P-H3-Image

Choose P-H3-Image when photoreal texture, film still energy, or open-weights style workflows matter more than MiniMax’s 16-reference hosted path.

Photo & Editorial

Documentary and Portrait Looks

Draft frames with natural skin, grain, and observed lighting instead of over-smoothed plastic texture.

Film & Concept

Cinematic Stills and Mood Plates

Block atmosphere-heavy stills that feel like single frames from a larger shoot.

Composite Edits

Multi-Picture Scene Builds

Combine several references into one dinner-party or product composite from a short instruction.

Graphic & Print

Short Quoted Text Accents

Use brief lettering sparingly; short quotes fare better than dense paragraphs.

Open Workflows

Diffusers / Comfy-Style Pipelines

Fit open inference stacks that want H3-derived quality with a smaller, faster package than full H3 video stacks.

Iteration

Fast Photoreal Drafts

Lean on eight-step sampling when you need high-detail stills without a long CFG dance.

How Creators Approach a Session in P-H3-Image

Whether drafting an 8 MP still or composing a multi-picture edit, keep the loop short and review-focused.

1
Start with a short request
Let the upsampler expand a clear one-liner, or disable it when you already have a full prompt.
2
Target the right megapixels
Prefer 8 MP for text-to-image; use 4–8 MP for edits when faces and fine detail matter.
3
Edit with named pictures
Picture 1 is the subject to change; later images are references the instruction can cite.
4
Review known failure modes
Check hands, tiny background figures, and any dense lettering before you ship.

P-H3-Image Prompt Ideas

Short requests work when the upsampler is on. For manual prompts, name the look, era, lens, and what must not change.

Editorial film still

T2I · 8 MP

1970s film still of a quiet coastal highway at dusk, warm sodium lamps, light film grain, medium telephoto compression, natural skin texture if people appear, no logos, no dense body copy.

Multi-picture composite edit

Edit · up to 8 images

Give the woman in picture 1 the hat from picture 2, keep her pose and face, soft window light, photoreal, no text.

Macro material study

T2I · 3:2

Extreme close-up of brushed titanium under a hard key light, fine machining marks, soft bounce from the left, museum-catalog clarity, no text.

Same H3 lineage. Different products.

Both sit in the MiniMax H3 family tree, but they are not interchangeable. MiniMax H3 Image is MiniMax’s own image generation and editing surface. P-H3-Image is Pruna’s open-weights model derived from MiniMax H3 technology.

Spec	MiniMax H3 Image	P-H3-Image
Publisher	MiniMax	Pruna AI (open weights)
Tuned for	Hosted text-to-image and multi-reference editing	Efficient open inference, photoreal and editorial stills
References	Up to 16 reference images	Up to 8 images (first edited, others as Picture 2…)
Detail target	1K · 2K · 4K area tiers / flexible sizing	Best around 8 MP text-to-image; edits at 4–8 MP
Prompting	Layered briefs and identity lock instructions	Built-in prompt upsampler; short requests expand well

Need MiniMax’s own H3 Image surface?

MiniMax H3 Image is MiniMax’s generation and editing product with up to 16 references and flexible 1K–4K area workflows. Choose it when you want the official H3 Image path rather than Pruna’s open-weights derivative.

Explore MiniMax H3 Image

Frequently Asked Questions About P-H3-Image

What is P-H3-Image?: P-H3-Image is Pruna AI’s open-weights image model based on MiniMax H3. It generates text-to-image stills and performs instruction edits with up to eight input images.
Is P-H3-Image the same as MiniMax H3 Image?: No. MiniMax H3 Image is MiniMax’s own image generation and editing surface. P-H3-Image is a separate open model derived from MiniMax H3 technology, with different reference limits and sizing guidance.
What resolution works best?: Text-to-image is strongest around 8 megapixels. Edits often look best at 4 or 8 MP. Lower resolutions can introduce more artifacts in faces, hands, and fine detail.
How do multi-image edits work?: Picture 1 is the image being edited. Additional images become Picture 2, Picture 3, and so on, so your instruction can borrow a hat, background, or prop from a named reference.
What is the prompt upsampler?: By default, short requests are expanded into fuller prompts before rendering. That helps conversational edits, but you can bypass it when you already have a complete prompt.
What are the main limitations?: Hype-heavy or awkward prompt lengths can weaken results. Hands and tiny background figures may distort, especially below 8 MP. Dense text is unreliable; short quoted strings work better. P-H3-Image does not generate video or audio.

Create with P-H3-Image

Explore how photographers, editors, and indie makers turn short requests into photoreal and cinematic stills with P-H3-Image.

Pair efficient 8 MP generation with multi-image instruction edits when a scene needs more than one reference.

Explore P-H3-Image

Explore MiniMax H3 Image: /minimax-h3-image