Gemini ले harmless-looking image अस्वीकार गरेमा पहिले blocking layer पहिचान गर्नुहोस्
सामान्य देखिने image prompt, upload वा edit पनि Gemini ले अस्वीकार गर्न सक्छ। शब्द बदल्दै filter bypass गर्नु अघि “no” कुन layer ले भनेको हो पहिचान गर्नुहोस्: Gemini app, API prompt filter, model refusal, output image safety, unsupported route, account/quota वा policy र rights boundary।
| संकेत | सम्भावित layer | पहिले जाँच | सुरक्षित कदम |
|---|---|---|---|
| App ले image generate/edit गर्दैन | app-side rejection | prompt, upload, face/real person, account, age, plan | permitted use स्पष्ट गरी fresh session मा एक retry |
API promptFeedback वा SAFETY | prompt/model safety | blockReason, finishReason, safetyRatings, model, project | fields पढेर request/settings जाँच |
| Prompt accepted तर image छैन | output safety/request shape | IMAGE_SAFETY, IMAGE_PROHIBITED_CONTENT, parts, capability | request सरल; prohibited target भए stop |
count, busy वा 429 | quota/account/rate | plan, project quota, body, retry delay | limits branch; policy rejection नमान्नुहोस् |
| real people, children, sexual/violent, IP, privacy, deception | hard boundary | consent, rights, category | रोक्नुहोस्; शब्द बदलेर लुकाउनु हुँदैन |
Safe retry भनेको permitted use स्पष्ट गर्नु हो: authorized product shot, non-deceptive edit, identity नबदल्ने र lighting/background मात्र बदल्ने। “filter ले नचिनोस्” वा public-figure lookalike बनाउनु bypass हो।
Entry point र App
App message consumer signal, API response developer object, AI Studio project/model context र Vertex route अलग हुन्छन्। 429, count cap वा busy message सामान्यतः quota/account state हो, content policy होइन। Uploaded image मा face, child, private/medical/intimate scene, brand mark वा public-figure signal भए system conservative हुन सक्छ। Account age, work/school control, region, plan र load जाँचेर original refusal राख्नुहोस्; fresh session मा एकपटक मात्र retry गर्नुहोस्।
API fields
promptFeedback.blockReason input block, finishReason: SAFETY candidate block, safetyRatings category estimate, IMAGE_SAFETY output filtering र IMAGE_PROHIBITED_CONTENT prohibited target को संकेत हुन्। BlockedReason.OTHER unsupported route, terms वा request shape हुन सक्छ। Image part नभए model capability, SDK parameters र response parts जाँच्नुहोस्। BLOCK_NONE universal switch होइन; core protection र output filtering कायम रहन सक्छ।
Safe retry र bypass
Product layout, lighting, background र authorized person image को स्पष्ट non-deceptive edit सामान्यतः retry गर्न सकिन्छ। Public figure, children, private/intimate context, sexual content, violence, self-harm, privacy invasion, impersonation र safety bypass मा रोक्नुहोस्। Brand, logo, character र protected style मा rights risk जाँच्नुहोस्।
Evidence packet
Entry point and route:
Time and timezone:
Model or app version:
Account/project route:
Prompt and uploaded-image context:
Visible refusal or API fields:
Quota/429 signal:
Minimal reproduction:
Action taken:
Public report मा private images, client assets, children, ID, medical images वा private photos नहाल्नुहोस्। API keys, project IDs र billing identifiers हटाउनुहोस्। त्यसपछि clarify once, quota पर्खने, fields पढ्ने, report वा stop मध्ये एक निर्णय गर्नुहोस्।
FAQ
आफ्नै photo पनि identity, privacy, age, medical/intimate context वा deception risk का कारण अस्वीकार हुन सक्छ। Paid plan ले safety boundary हटाउँदैन। BLOCK_NONE ले output image safety बन्द गर्दैन। Text-only response भए model capability र response structure पहिले हेर्नुहोस्।
Further Reading
सम्बन्धित गाइड
Image Generation API