Making pictures with AI on your own computer

Twelve picture models you can download and run at home, each with its licence, the memory it asks for and its download size. The model runs on your computer, not on a server.

  • Two model families
  • A fit guide for each model
  • 12 Picture models in the table
  • 4 GB Memory the smallest ones ask for

Stable Diffusion 1.5 models, and SD-Turbo

Fits on is the memory each model asks for, a fit guide and not a limit.

Stable Diffusion 1.5

The base SD 1.5 model. It makes a picture from words and repaints a picture you have. It comes as the full file or a lighter fp16 one.

  • Licence: CreativeML OpenRAIL-M
  • Fits on: 4 GB memory, 2 GB graphics memory
  • Download: 4.27 GB, or 1.99 GB as fp16

DreamShaper 8

An SD 1.5 model, and the one we have timed: 512x512 in about 13 s on a recent Mac, in our tests on a pre-release build.

  • Licence: CreativeML OpenRAIL-M
  • Fits on: 4 GB memory
  • Download: 1.99 GB

Realistic Vision V6.0 B1

An SD 1.5 model, saved in half precision (fp16), so the file is about half the size of a full one.

  • Licence: CreativeML OpenRAIL-M
  • Fits on: 4 GB memory, 2 GB graphics memory
  • Download: 2.13 GB

AbsoluteReality v1.8.1

An SD 1.5 model. Pictures from words, and repainting a picture you have.

  • Licence: CreativeML OpenRAIL-M
  • Fits on: 4 GB memory
  • Download: 1.99 GB

LCM DreamShaper v7

A fast SD 1.5 model: an LCM version of DreamShaper, made to finish a picture in a few steps.

  • Licence: CreativeML OpenRAIL-M
  • Fits on: 4 GB memory
  • Download: 3.2 GB

SD-Turbo

Built on Stable Diffusion 2.1 and made to draw a picture in one step.

  • Licence: Stability AI Community License
  • Fits on: 6 GB memory
  • Download: 4.86 GB

SDXL models

SDXL models ask for 6 to 8 GB of memory, more than the SD 1.5 ones.

  1. SDXL Base 1.0

    The base SDXL model. It makes a picture from words and repaints a picture you have.

    • Licence: CreativeML OpenRAIL++-M
    • Fits on: 8 GB memory, 6 GB graphics memory
    • Download: 6.94 GB
  2. SDXL-Turbo

    An SDXL model made to draw a picture in one step, saved in half precision (fp16).

    • Licence: Stability AI Community License
    • Fits on: 8 GB memory
    • Download: 6.46 GB
  3. Segmind SSD-1B

    A distilled SDXL: a smaller model that asks for less memory than the other SDXL rows.

    • Licence: Apache 2.0
    • Fits on: 6 GB memory
    • Download: 4.16 GB
  4. RealVisXL V5.0

    An SDXL model, saved in half precision (fp16).

    • Licence: CreativeML Open RAIL++-M
    • Fits on: 8 GB memory, 6 GB graphics memory
    • Download: 6.94 GB
  5. Juggernaut XL v9

    An SDXL model. Its licence adds a RunDiffusion addendum to the OpenRAIL-M terms: read it in the app before you accept.

    • Licence: CreativeML OpenRAIL-M (RunDiffusion addendum)
    • Fits on: 8 GB memory, 6 GB graphics memory
    • Download: 7.1 GB
  6. DreamShaper XL Lightning

    An SDXL Lightning model, made to finish a picture in a few steps.

    • Licence: CreativeML Open RAIL++-M
    • Fits on: 8 GB memory, 6 GB graphics memory
    • Download: 6.94 GB

What the names and settings mean

Four things that explain most of the difference between these models.

SD 1.5, SDXL and one-step models

SD 1.5 models are the small family: 4 GB of memory in the table above. SDXL models are the bigger one: 6 to 8 GB. Turbo, Lightning and LCM models are trained to finish in very few steps, down to one for SDXL-Turbo. Fewer steps is faster, with less detail.

Steps, size and seed

A picture model starts from noise and cleans it up a little at each step. More steps take longer. Size is the picture in pixels: a bigger picture needs more memory and more time. The seed is the number the noise starts from: the same seed, words and settings give the same picture again on the same setup.

Repaint: picture to picture

Give the model a picture and a sentence, and it starts from your picture instead of from noise, so the layout stays. In Pineido Studio this is how you restyle a picture, add or remove things, or change colours. There is no mask: it works on the whole picture.

Why you accept a licence first

Most of these models use an OpenRAIL licence. It lets you use the model, and it lists uses you must not make; if you pass the model on, those rules go with it. The two Turbo models use the Stability AI Community License, with terms of its own. SSD-1B uses Apache 2.0.

Models that understand a photo

These describe a photo you give them, and some read the text in it.

  1. InternVL3.5-14B

    Describes a photo. It described a photo correctly in about 100 s on a 48 GB Mac, in our tests on a pre-release build.

    • Licence: Apache 2.0
    • Fits on: 14 GB memory, 10 GB graphics memory
    • Download: 9 GB
  2. InternVL3.5-2B

    Small and fast. In our tests on a pre-release build it described photos and read the text in a picture. It can misread some Cyrillic lines.

    • Licence: Apache 2.0
    • Fits on: 4 GB memory, 2 GB graphics memory
    • Download: 1.92 GB
  3. Tesseract English text reading

    Reads English text in a picture when no vision model is installed.

    • Licence: Apache 2.0
    • Comes on the stick: no download

How Pineido Studio handles picture models

Pineido Studio, our AI app on a USB stick, offers every model in this table as a download.

No model on the stick

No model comes on the stick. You pick one and the app downloads it.

The licence comes first

The app shows the licence and asks you to accept it before download when the licence requires it.

A fit label for your computer

The fit label tells you if a model fits your computer. A model that's too big still runs, slower, spilling to the processor.

Words in, picture out

Make a picture from words, or change one you have from a sentence.

Common questions, plain answers

Which picture model should I start with?

Start small. An SD 1.5 model such as DreamShaper 8 downloads at about 2 GB and asks for 4 GB of memory. It is the one we have timed: 512x512 in about 13 s on a recent Mac, in our tests on a pre-release build. Try an SDXL model when your computer has 8 GB of memory to give it.

Do I need a graphics card to make pictures?

No, but it decides the speed. Picture speed depends on your graphics chip; on a computer without one, a picture can take many minutes.

What does "fits on" mean in the table?

It is the memory the model asks for: a fit guide, not a limit. In Pineido Studio a model that is too big still runs, slower, spilling to the processor, and the fit label tells you so before you start.

Can I use the pictures in my business?

Read the licence of the model you pick: that is where the answer is. The OpenRAIL licences on most rows list uses you must not make. The two Turbo models use the Stability AI Community License, which has terms of its own. SSD-1B uses Apache 2.0. Pineido Studio shows the licence before the download.

Can it read the text in a photo?

Yes. InternVL3.5-2B read the text in a picture in our tests on a pre-release build, though it can misread some Cyrillic lines. When no vision model is installed, Pineido Studio reads English text with Tesseract, which comes on the stick.

Does Pineido Studio come with these models?

No. No model comes on the stick. You pick one in the app and it downloads it, after you accept the licence when the licence requires it.

12 Picture models in the table

Want pictures made on your own machines?

Tell us what pictures you need and which computers you have.

For one computer, Pineido Studio does this on a USB stick.

You leave with

  1. Tell us the pictures you need
  2. We test models on your computers

ByFounder and Imaginator

Pineido, based in Strumica, North Macedonia, is a software company that builds websites, online shops, web and mobile apps, business automation and AI for companies at home and abroad. Work ships in 30-day sprints at a fixed price, and the client owns the code.

Published