Signal / dreamgen-review

DreamGen review: Lucid roleplay models, credit plans, and privacy

DreamGen runs 2 of its own Lucid fine-tunes for roleplay and stories, on credit plans from $7.83/mo with an OpenAI-compatible API. What each plan buys and what the privacy policy says.

DreamGen is a roleplay and story-writing platform based in Switzerland. It runs its own Lucid models, fine-tuned for fiction, in a web app and through an OpenAI-compatible API that works from front-ends like SillyTavern. We list 2 models, both fine-tuned.

This review is based on DreamGen’s published plans, model list, and policies, checked on October 11, 2026. Benchmark results from our own runs will be added when they’re complete. Live details are on the DreamGen provider page.

The short version

Best for Roleplay and collaborative story-writing, in its app or from SillyTavern
Uncensored models 2 (both fine-tuned)
Per-token price None published: credit-based plans
Plans Free with limits; Starter $7.83/mo, Advanced $19.35/mo, Pro $48.30/mo
Privacy score 28 / 100, joint 26th of 37
Watch out for Content is auto-moderated by default, and the web app may route to unnamed third parties under load

Models

DreamGen’s two Lucid models:

  • lucid-v1-medium: a 12B fine-tune of the Mistral Nemo base model. Its weights are on Hugging Face (as lucid-v1-nemo) under a non-commercial license, so you can also run it yourself.
  • lucid-v1-extra-large: a 70B fine-tune of Llama.

Both are trained for roleplay and story-writing rather than uncensored by removing refusals. One thing to know up front: DreamGen’s content is auto-moderated by default. Its models are built for mature fiction, but you shouldn’t assume the default settings are unfiltered. We don’t have a published context window for either model, beyond what each plan allows. Search them on the Models page.

Pricing

DreamGen doesn’t publish per-token prices, so we can’t give a cost per 1,000 chats. You buy a monthly plan that gives you credits, and plans differ mainly in credits and context (how much of the story the model can see at once).

  • Free: the Lucid models can be used free, with limits.
  • Starter, $7.83/mo: 375 credits a month plus 45 extra a day, with a 5,000-token context.
  • Advanced, $19.35/mo: 2,250 credits a month plus 75 a day, with a 15,000-token context.
  • Pro, $48.30/mo: the same 2,250 monthly and 75 daily credits as Advanced, a 30,000-token context, 20% off credit packs, and up to 60% off select third-party models.

The pricing page was also showing first-month discounts (20% off Starter and Advanced, 30% off Pro) when we checked. The list prices above are the ones that recur.

Two things to read carefully. First, Pro doesn’t give you more credits than Advanced. You’re paying the extra $28.95 for double the context and the discounts. Second, we couldn’t find how many tokens a credit buys, so it’s hard to compare a DreamGen plan with a per-token API. If your stories are long, context is the number that matters most, and 5,000 tokens on Starter is short for ongoing roleplay.

Privacy

DreamGen scores 28 / 100 on our privacy rubric, which only gives points for documented claims:

Rubric item DreamGen Points
Prompt retention The web app stores your stories 0 / 20
No training on prompts Training on your content is opt-in 15 / 15
Technical guarantee (TEE / E2EE) No 0 / 20
Signup Account required 3 / 8
Crypto payments Payment methods not stated 0 / 7
Jurisdiction Switzerland 10 / 10
Upstream hosts disclosed No: may route to unnamed third parties under load 0 / 10
Independent audit No 0 / 10

That’s joint 26th of 37. The app keeps your stories, which is the point of a writing app but costs it the retention points. Training being opt-in is a genuine strength, and Switzerland is a clear jurisdiction.

There’s a split worth knowing. DreamGen’s model card says the API doesn’t store or log inputs or outputs, which is better than the app. Our score covers the service as a whole, so it uses the app’s policy. If privacy matters to you, the API from your own front-end looks like the better route on paper. The other caveat is in the policy itself: under load, the web app may send requests to third-party providers it doesn’t name.

Access and payment

  • Web app and an OpenAI-compatible API (it works from SillyTavern).
  • Account required.
  • Payment methods aren’t stated on the pages we checked.
  • Based in Switzerland.

How it compares

  • Unlimited instead of credits: NovelAI is $10/mo for unlimited text, and $25/mo for its 355B model with a 28,672-token context.
  • Flat-rate API hosting: Infermatic is $9/mo for API access to roleplay fine-tunes up to 72B with 32K context or more, and scores 40 on privacy.
  • Free: Agnai has free guest roleplay models with no account. AI Horde is free and volunteer-run, though volunteers can technically see prompts.

Verdict

DreamGen is worth a look if you want fiction-trained models from a Swiss company that doesn’t train on your writing unless you opt in. The free tier lets you try Lucid first, and the smaller model’s weights are public.

The plans are harder to judge. Credits aren’t converted to tokens, Pro and Advanced share the same credit allowance, and Starter’s 5,000-token context is tight for long stories. Check the moderation settings before you start.

See DreamGen’s models and plan details on its provider page, or compare every provider in the index.