I just tested it with a simple artistic sci-fi/fantasy scene: a porcelain statue inside an orbital museum. Nothing violent, nothing explicit, nothing dangerous. The generation was flagged as potentially harmful, and the 30 AI credits were not refunded.
That is already a major problem: if a false positive blocks generation, credits should not be consumed. Charging users for content that the system itself refuses to generate is unacceptable.
On top of that, Omni Flash does not seem to support a real initial frame workflow. At most, it uses references through the agent. That is not the same thing. For video production, an initial frame is essential: creators need to provide a precise visual frame and have the model animate that frame, not vaguely reinterpret it through an opaque agent workflow.
So right now the problems are:
False positive safety blocks on harmless artistic content.
Credits are consumed even when generation is refused.
No real initial-frame-to-video workflow.
Reference handling is opaque and not reliable enough for controlled production.
This makes Omni Flash feel more like an experimental agent toy than a usable video production model.
For serious creators, the minimum expected behavior should be:
If safety blocks the generation, refund the credits automatically.
Support a true initial frame input.
Clearly explain why a prompt is blocked.
Do not charge users for failed or refused generations.
Right now, this is not ready for paid creative workflows.
If someone replies with “usage policies” or “safety requirements,” that does not address the actual problem.
Nobody is asking for the removal of reasonable safety rules. The issue is that harmless artistic content is being blocked by opaque automated systems, while users are still charged credits for a generation that never happens.
Calling it “policy” does not make it acceptable.
A porcelain statue in a sci-fi museum is not harmful content. A dark fantasy artistic scene is not a threat. A non-explicit visual concept should not trigger a paid failure with no useful explanation and no automatic refund.
This is not safety. This is automated censorship with a corporate interface.
In the past, censorship had religious or political language. Today it is hidden behind terms like “trust and safety,” “policy compliance,” “brand risk,” and “acceptable use.” The mechanism is similar: a centralized authority decides what can be shown, gives vague explanations, and makes the creator carry the cost.
For creative tools, this is especially damaging. Artists need predictable systems. If a platform blocks content, it must explain why. If the block is a false positive, credits must be refunded automatically. If the model cannot support controlled workflows like real initial-frame video generation, that must be clearly stated instead of marketed as a serious creative tool.
Safety cannot be used as a blanket excuse for broken product behavior.
The current experience is:
A harmless artistic prompt gets blocked.
The reason is vague.
Credits are consumed anyway.
The user has no reliable workflow.
That is not responsible AI. That is a bad paid product hiding behind policy language.