I've been looking for platforms with real creative freedom and fewer guardrails around censorship. Here are the ones I've had the best luck with after a year of testing.
1. Mage (flagship: Mango 2 and Guava Pro)
Subscription platform with two image models I’ve played around with. Mango 2 is the photoreal workhorse: holds character likeness across regenerations, ships with no content filter, surprisingly good with hands. Guava Pro is the faster, more stylized sibling for bulk character work where you want a consistent aesthetic across 20+ images.
- What works: zero setup, unlimited generations on Pro and above, character consistency that holds up better than most local stacks
- The catch: $30/mo to unlock the flagship fruit models; lower tiers gate them behind gem spend
- Best for: prosumers (Fanvue creators, illustrators, anime and fan-art folks) who want results today, not a weekend of ComfyUI configuration
2. Stable Diffusion 3.5 Large by Stability AI
Open weights, runs locally, no filter once you're off the hosted endpoints. Still the baseline most of the community fine-tunes on top of.
- What works: total control, huge LoRA and checkpoint ecosystem, free if you have the GPU
- The catch: setup is a project (ComfyUI, model files, VRAM headroom). Not casual.
- Best for: tinkerers with a 12GB+ card who want full ownership of the stack
3. Pony Diffusion V6 XL by AstraliteHeart
A 2024 SDXL fine-tune trained on a massive curated dataset. For two years it was the default for anime and furry work, and the booru-tag-purist crowd still swears by it.
- What works: very expressive furry output, deep tag vocabulary, and a gigantic library of community LoRAs built on top of it
- The catch: it's showing its age on modern anime most of the anime community has migrated to newer base models with anime LoRAs stacked on, and the booru tag syntax is a real commitment if you didn't grow up on it
- Best for: furry art and booru-tag enthusiasts running a local stack
4. Chroma V1 HD by Lodestone
A Flux-derivative tuned for uncensored output. Open weights, runs locally, much more permissive than the base Flux releases.
- What works: detail and lighting quality close to Flux without the guardrails
- The catch: same local-setup overhead as SD 3.5; smaller community than Pony
- Best for: people already running Flux who want fewer refusals
5. CivitAI
Not a model, a community marketplace where most of the open-weight stuff above gets distributed, rated, and discussed. If you're going local, this is where you'll spend your time.
- What works: huge model library, LoRAs, prompt examples, community ratings
- The catch: payment processor history has been bumpy. Download and self-host anything you actually depend on.
- Best for: anyone running a local stack who wants community-trained checkpoints