Local-tool capabilities
The denominator is the 18 entities in the canonical Local AI Tools category. All 18 retain the Free / local / open-source access classification formerly visible in the combined access dataset.
Local tool comparison
| Entity | Modalities | Access model | API status | License | Hardware note |
|---|---|---|---|---|---|
| ComfyUI | Image generation, Video generation, Audio generation, 3D generation | Free, open-source | API documented | GNU GPL-3.0 | NVIDIA GPU recommended; AMD ROCm, Intel Arc, and Apple Silicon also supported; low-VRAM configurations and CPU fallback are possible but slower |
| AUTOMATIC1111 / Stable Diffusion WebUI | Image generation | Free, open-source | API documented | GNU AGPL-3.0 | NVIDIA GPU recommended; README cites 4GB VRAM as workable (with community reports of 2GB), 8GB+ recommended for training |
| Stable Diffusion WebUI Forge | Image generation | Free, open-source | Not documented | GNU AGPL-3.0 | NVIDIA GPU recommended; includes a GPU memory-management system intended to reduce VRAM pressure versus the base WebUI, though no quantified benchmarks are published |
| Fooocus | Image generation | Free, open-source | Not documented | GNU GPL-3.0 | 4GB VRAM minimum on Nvidia GPUs (8GB on AMD); 8GB system RAM minimum plus swap space; CPU-only mode supported but roughly 17x slower than GPU |
| InvokeAI | Text-to-image, Image-to-image, Inpainting, Outpainting, Upscaling | Free, open-source | Not documented | Apache License 2.0 | NVIDIA GPU (Windows/Linux) or AMD GPU (Linux only, via ROCm) recommended; Apple Silicon supported; VRAM needs scale by model (roughly 4GB for SD1.5 up to 24GB for larger FLUX variants); CPU-only mode available on some platforms |
| SillyTavern | Text chat, Image generation (via connected backend), Text-to-speech | Free, open-source | No public API documented | GNU Affero General Public License v3.0 (AGPL-3.0) | Runs on any device that supports Node.js 20+; hardware needs are determined by whichever local LLM backend is connected, if any |
| KoboldCpp | Text generation, Image generation (Stable Diffusion), Speech-to-text (Whisper), Text-to-speech, Image recognition | Free, open-source | API documented | GNU Affero General Public License v3.0 (AGPL-3.0); built on the MIT-licensed llama.cpp | Runs on CPU alone or with GPU acceleration via CUDA (NVIDIA), Vulkan (any GPU), or HIP/ROCm (AMD); VRAM needs scale with model size and quantization level, with layer offloading configurable to fit available VRAM |
| LM Studio | Text chat, Local document retrieval (RAG) | Free for personal and business use | API documented | Not publicly documented | macOS 14+ on Apple Silicon (Intel Macs not supported); Windows needs AVX2 support on x64; 16GB+ RAM recommended across platforms; at least 4GB dedicated VRAM recommended on Windows for GPU acceleration |
| RisuAI | Text chat, Text-to-speech, Character media assets | Free open-source client; connected inference may be local, free, subscription, or usage-priced | API documented | GPL-3.0 | Client requirements are modest; local model hardware depends on the chosen inference backend |
| Agnai | Text chat, Third-party image generation | Free open-source self-hosting and public hosted service; inference-provider costs vary | API documented | AGPL-3.0 | Frontend can use remote APIs; local model hardware belongs to the connected backend |
| FaceFusion | Image editing, Video editing, Face swap, Lip sync | Source-available local software | API documented | Project-specific license; bundled model licenses and restrictions must be reviewed separately | CPU supported; documented accelerator paths include CUDA, CoreML, DirectML, MIGraphX, and OpenVINO |
| SwarmUI | Text-to-image, Image-to-image, Image editing, Text-to-video, Image-to-video, Upscaling | Free open-source software | API documented | MIT | Depends on model family, precision, quantization, resolution, frames, and backend offloading |
| SD.Next | Text-to-image, Image-to-image, Image editing, Image and video processing, Captioning | Free open-source software | API documented | Apache-2.0 | Supports multiple CPU/GPU backends; requirements vary by model, precision, offload, and resolution |
| Stability Matrix | Package management, Model management, Image generation through installed packages | Free application with source available under documented license terms | No public API documented | AGPL-3.0 source; distributed binaries have separate documented terms | Manager is lightweight relative to installed generators; each package and model has its own GPU/RAM requirements |
| text-generation-webui | Text generation, Text chat, Vision input, Image generation | Free open-source software | API documented | AGPL-3.0 | CPU-only and GPU backends supported; practical model size depends on RAM, VRAM, format, and quantization |
| llama.cpp | Text generation, Text chat, Embeddings, Multimodal input where model-supported | Free open-source software | API documented | MIT | CPU inference supported; GPU acceleration and CPU/GPU hybrid offload supported across multiple backends |
| Ollama | Text generation, Text chat, Embeddings, Vision where model-supported | Free local software; optional hosted capabilities may have separate terms | API documented | MIT for the core repository; individual model licenses vary | Depends on model size, quantization, context, and available CPU/GPU memory |
| Kohya_ss GUI | LoRA training, DreamBooth training, Diffusion fine-tuning, Training sample generation | Free open-source software | API documented | Apache-2.0 | Training-capable GPU strongly preferred; practical VRAM depends on base model, resolution, precision, optimizer, and caching |
Methodology
Inclusion uses the canonical Local AI Tools type. Modalities, API, license and hardware notes are displayed from structured profile fields; no unlisted format or operating mode is inferred.
Unknown handling: Missing or insufficiently specific fields remain Not documented and are never counted as No. A documented modality establishes support; quality and feature parity are not inferred.
Limitations
The category is a curated selection, not a complete inventory. Hardware needs vary substantially by model, quantization, resolution and workflow, so profile notes are not benchmark results.
No historical trend is shown because this release has one comparable snapshot.
Technical context
This count spans image and language stacks with different formats, memory scaling and execution layers.
Local Diffusion Inference StacksThe loaders, execution engines, precision choices, memory strategies, interfaces, and extensions behind local image generation.Local LLM Inference StacksHow model weights, GGUF quantization, execution backends, chat templates, context caches, and front ends combine for local character chat.Stable Diffusion XL (SDXL)A larger latent-diffusion family with dual text encoders, multi-aspect training, and an optional refinement stage.FLUX.1Black Forest Labs’ rectified-flow transformer family spanning open-weight, locally runnable and hosted image models.
Related concepts
Related Blog posts
How to Run Adult AI Image Generation LocallyGuidance that uses these fields and classifications.How Much VRAM Do You Need for Local AI Generation?Guidance that uses these fields and classifications.ComfyUI, AUTOMATIC1111, Forge, Fooocus or InvokeAI: Choosing a Local InterfaceGuidance that uses these fields and classifications.
Other study
Sources and provenance
- AI Porn Database canonical entity roster and profiles, snapshot 2026-08-11
- Official platform/company documentation retained in each linked Database profile’s Sources section.
- Normalized snapshot:
tools/statistics_data/2026-08-11.json; normalization version 2026-08-10.2.