Answer · confidence: partial
Who made downloadable AI for everyone, and what guides helped run it at home?
Research question: Who prepared the first downloadable AI models for ordinary people, and what guides or tools helped them get the models running at home?
The records show that Stability AI released Stable Diffusion as an open-source text-to-image model with publicly available weights in August 2022 2. Its code and weights were released publicly, and an optimized version can run on most consumer hardware with a modest GPU needing as little as 2.4 GB of VRAM 1. Earlier, EleutherAI's Katherine Crowson and Ryan Murdock adapted image generation models into text-to-image models using CLIP 8, and Crowson released VQGAN-CLIP by tweeting notebooks that let people run it for free without special equipment 10. PCWorld described Fooocus as the easiest way to create AI art on a PC 7, and tools such as Ollama serve quantized models so end users can run 1B–14B models on 4–16GB of VRAM 12. The verified facts do not identify who was first to offer downloadable models to ordinary people, so this answer is only partial 3, and the newest fact is dated 2026, within the current year 9.
Checked 2026-10-21. Only documented and supported facts are used.
Dig deeper on this questionSources
[1] supported
Stable Diffusion's code and model weights were released publicly, and an optimized version can run on most consumer hardware with a modest GPU needing as little as 2.4 GB of VRAM.
"Its code and model weights have been released publicly , [ 8 ] and an optimized version can run on most consumer hardware equipped with a modest GPU with as little as 2.4 GB VRAM ."Stable Diffusion
[2] supported
Stability AI released Stable Diffusion as an open-source text-to-image model with publicly available model weights.
"August 22 – Stability AI releases Stable Diffusion , an open-source text-to-image model with publicly available model weights."2022 in artificial intelligence
[3] supported
This public release marked a departure from earlier proprietary text-to-image models such as DALL-E and Midjourney, which were only accessible through cloud services.
"arture from previous proprietary text-to-image models such as DALL-E and Midjourney , which were accessible only via cloud services ."Stable Diffusion
[4] supported
The open-source text-to-image model Stable Diffusion was first released in August 2022.
"open-source text-to-image model Stable Diffusion was first released"Open-source artificial intelligence
[5] supported
Meta AI released OPT-175B, a 175-billion-parameter language model, and made it available to the research community.
"May 3 – Meta AI releases OPT-175B, a 175-billion-parameter language model trained on publicly available datasets and made available to the research community."2022 in artificial intelligence
[6] supported
The release of Meta's Llama models was a milestone in generating interest in open-weight and open-source models.
"The release of the Llama models was a milestone in generating interest in open-weight and open-source models."Open-source artificial intelligence
[7] supported
PCWorld described Fooocus as the easiest way to create AI art on a PC.
"Fooocus is the easiest way to create AI art on your PC"Stable Diffusion
[8] supported
EleutherAI's Katherine Crowson and digital artist Ryan Murdock developed a technique that adapted regular image generation models into text-to-image models using CLIP, a model developed by OpenAI.
"EleutherAI's Katherine Crowson and digital artist Ryan Murdock developed a technique for using"EleutherAI
[9] supported
Model compression techniques and mixture-of-experts architectures have lowered hardware requirements so that 7- to 33-billion-parameter models can run on consumer-grade workstations.
"model compression techniques (knowledge distillation, structured pruning, and low-bit quantization) together with mixture-of-experts architectures have lowered hardware thresholds enough that 7- to 33-billion-parameter models now run on consumer-grade workstations with modest neural processing units or graphics processing units."Local Deployment of Open-Weight Language Models in Dermatology: Viewpoint on Privacy, Equity, and Practical Implementation (2026)
[10] supported
Katherine Crowson released the VQGAN-CLIP technology by tweeting notebooks, which let people run the technique for free without any special equipment.
"Crowson released the technology by tweeting notebooks demonstrating the technique that people could run for free without any special equipment."EleutherAI
[11] supported
EleutherAI released GPT-J-6B on June 9, 2021, a six billion parameter model that was then the largest open-source GPT-3-like model in the world.
"On June 9, 2021, EleutherAI followed this up with GPT-J-6B , a six billion parameter language model that was again the largest open-source GPT-3-like model in the world."EleutherAI
[12] supported
Tools such as Ollama serve quantized models, compressing weights to about 2.6 bits per weight so end users can run 1B–14B models on 4–16GB of VRAM.
"Served through tools such as Ollama, it compresses weights from 16-bit precision (FP16, BF16) to as little as ∼2.6 bits per weight (Q2_K), which allows the end user to run 1B–14B models within 4–16GB of VRAM"Saying, Not Knowing: Aggressively GGUF-Quantized Small Language Models Still Write Rare Words They Can No Longer Define (2026)
[13] supported
EleutherAI's language models were released under the Apache 2.0 free software license, which made them freely usable.
"These language models were released under the Apache 2.0 free software license"EleutherAI