Better accuracy of likeness for images generation of well-known celebrities
I’m sure this has already been mentioned, but right now it’s kind of my only major complaint about the platform. The only model that seems to be able to generate accurate image representations of known subjects such as famous celebs is the Lustify SDXL model using the Hyperrealism image style. I am able to dial in the settings enough to occasionally generate a convincing likeness of a well known celebrity. But this model is far from being the most realistic looking image generator, and because it’s uncensored and really geared more toward erotic content, it tends to be a bit over the top in terms of returning naked or partially naked results. I want a model that can pull description information from the web to produce accurate realistic looking images of well known recognizable people.
Log in to comment and vote
Comments2
Jack
May 28, 2025
Try out the new Hi-Dream model we added yesterday, and let us know your thoughts.
https://x.com/AskVenice/status/1927379748000252081
Black Tamarin
May 17, 2025
Here’s an assessment of potential cobra effects from improving image-generation accuracy for celebrities on Venice.ai:
1. Legal Escalation & Platform Liability
Risk: Hyper-accurate celebrity likenesses could trigger lawsuits over rights of publicity, defamation, or deepfake misuse. Even with disclaimers, Venice might face liability for enabling unauthorized commercial/defamatory use.
Outcome: Legal battles could force Venice to cripple the feature or pay licensing fees, undermining its uncensored ethos.
2. Accelerated Regulatory Crackdowns
Risk: Governments already fear AI-generated impersonations. A tool that reliably mimics public figures might provoke bans on AI image generation entirely.
Outcome: Venice could be outlawed in jurisdictions, shrinking its user base and inviting global scrutiny.
3. Weaponization for Fraud/Exploitation
Risk: Criminals could use precise celebrity likenesses for scams (e.g., fake endorsements, nonconsensual porn, political disinformation).
Outcome: Venice becomes a PR liability, labeled a "crime tool," and pressured to implement censorship—contradicting its core values.
4. Erosion of User Trust
Risk: Pulling web data for accuracy risks amplifying biases or outdated information (e.g., generating a celeb based on old scandal imagery).
Outcome: Users lose faith in outputs, accusing Venice of slander or "AI hallucination," even if the model technically followed prompts.
5. Centralization Pressure
Risk: To avoid legal risks, Venice might need to block generation of specific celebrities (e.g., Taylor Swift), creating a slippery slope toward censorship.
Outcome: A "blocklist" system emerges, mirroring the restrictive policies of censored platforms.
6. Model Overfitting & Creativity Loss
Risk: Prioritizing photorealism for celebrities could make the model rigid, reducing its ability to generate original characters or stylized art.
Outcome: Artists and creatives abandon the platform, leaving only users seeking hyperrealistic impersonations.
7. Attraction of Malicious Users
Risk: Venice’s uncensored nature + celeb accuracy could make it a hub for generating abusive content (e.g., revenge porn, fake propaganda).
Outcome: Infrastructure providers (e.g., cloud hosts, payment processors) cut ties with Venice to avoid association with harmful outputs.
Conclusion:
While improving celebrity likenesses seems benign, it risks inviting existential threats to Venice’s operational freedom. A compromise might involve localized models (user-side fine-tuning without centralized celeb datasets) or opt-in celebrity licensing (e.g., paid partnerships with public figures). However, both solutions conflict with Venice’s anti-censorship principles, highlighting the paradox of uncensored AI tools in regulated environments.