popnow
We pay you backProtectedPricingSign inStart creating

All models

Pruna AI AI models

3 models · from 6 credits

At a glance

Models
3
Cheapest to try
P-Image LoRA · 6 credits
Newest
P-H3-Image LoRA · November 3, 2026
Types
Image

The models

Choosing a Pruna AI model for fast drafts, large stills or your own LoRA

Three image models from one lab

Pruna AI builds fast versions of open image models, and PopNow offers three of them. P-Image LoRA is the speed-focused one: it makes an image from text in seven shapes or a custom size up to 1440 pixels a side, edits with up to five source pictures, and loads one LoRA of your choice. P-H3-Image is Pruna's still-image model adapted from the MiniMax H3 open weights and released as open source in November 2026. It writes a picture from a described scene or edits one using up to nine pictures, in five sizes from 1K to 4K and fifteen aspect ratios. P-H3-Image LoRA is the same model with up to three MiniMax H3 LoRA files loaded, each with its own strength from 0 to 4.

Size, time and how the price grows

P-Image LoRA returns images in a few seconds, which suits quick drafts and trying a style on a real prompt. P-H3-Image is slower but goes much larger: a 1K image usually takes well under a minute, while a 4K image of about 16.8 megapixels takes around four minutes, so try an idea at 1K first and make the final at 3K or 4K. On P-H3-Image the price rises with the size tier, an edit adds a small charge for each picture you add, and the LoRA version adds a flat amount per image. The Generate button always shows the exact price before you start, and the page for each model lists the current rates.

Using your own LoRA

Pick the model that matches how your LoRA was trained. P-Image LoRA loads one LoRA made for P-Image for text to image, and LoRAs made for Qwen Image Edit when editing. P-H3-Image LoRA loads up to three LoRAs trained for MiniMax H3, which makes it a good fit for a character or house style you already use with H3. Paste the link that downloads the .safetensors file itself, such as a Hugging Face resolve link, not the page that describes the model. On P-H3-Image LoRA a LoRA the model cannot load is skipped while the image is still charged, so PopNow checks each link before the run starts and asks you to fix it if it is not a downloadable .safetensors file.

Quick answers

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

No. Pruna AI adapted the MiniMax H3 open weights into a separate still-image model, so the same prompt and seed give a different picture from MiniMax H3 Image LoRA, which runs MiniMax's own weights at 1K or 2K. Choose P-H3-Image when you want 3K or 4K output or edits from up to nine pictures.

Which Pruna model is the fastest?

P-Image LoRA is built for speed and usually returns an image in a few seconds. P-H3-Image takes longer and depends on the size you pick: 1K usually finishes well inside a minute, while 4K takes about four minutes. Draft quickly, then switch sizes for the final image.

How do I point to a picture in a P-H3-Image edit?

Put the picture whose scene you want to keep first and call it <Picture 1> in your prompt. The next pictures are <Picture 2>, <Picture 3> and so on, so you can say which person, outfit or object should come from which picture. Without a chosen ratio, the edit follows the first picture.

Why was my LoRA link not accepted?

The link probably opens a web page or a file that is not a .safetensors LoRA. P-H3-Image LoRA skips a LoRA it cannot load but still charges for the image, so PopNow checks the link first. Use the direct download link to the .safetensors file, such as a Hugging Face resolve link.