When working with neural networks, it is important to follow established rules and guidelines in order to comply with ethical standards and legal requirements.
At SYNTX.AI, different types of neural networks are used: from models without built-in filters, which provide almost complete freedom of generation, to models with strict moderation systems, where violations of the rules may lead to access restrictions or account blocking.
Filters are built-in restrictions that may cause a neural network to reject a request and not generate a result.
Most often, filters are triggered by:
- adult content
- scenes of violence
- political topics
- public figures
- brands and copyrighted materials
Understanding how filters work helps you choose the right model and avoid generation refusals.
- Neural Networks with Strict Filters
- Neural Networks with Partial Filters
- Neural Networks with Minimal Filters
- Additional Group of Tools
- Important to Know
Neural Networks with Strict Filters
This category includes the following models:
- 𝐓 Ideogram v3.0
- 🌙 SORA
- 🌙 SORA (GPT) Image
❌ Prohibited Content
- Violations of Rules and Laws
- use of copyrighted materials
- fraud, spam, or deception of users
- extremist content
- Dangerous and Sensitive Content
- erotic or pornographic content
- violence and cruelty, including cruelty toward animals
- suicide and self-harm
- discrimination, racism, hate speech, or insults
- Content Involving People
- use of public figures in sensitive or inappropriate contexts
- scenes involving children, especially in ambiguous or potentially unsafe situations
4. Scene and Context Restrictions
- images of people in intimate or private situations, such as in the shower or bathroom
- indecent, overly explicit, or suggestive content
- сertain words and phrases may also trigger filters, for example: “wet clothing,” “cleavage,” “breast,” “underwear,” “latex,” “nude,” “skin,” “girl,” “boy,” “waist,” “body,” etc.
❗ Additional Notes
Many models may reject requests involving children or specific age references in order to prevent the generation of potentially inappropriate content.
Filters often respond not only to individual words, but also to the overall meaning and context of a request. Even neutral wording may lead to a refusal if the system interprets the request as potentially violating its rules.
Neural Networks with Partial Filters
These neural networks have less strict filters compared to the previous category of tools.
They are generally less likely to fully block a request, but restrictions still apply and may depend on the wording, context, and overall meaning of the prompt.
This category includes the following neural networks:
- 🧿 Seedance
- 🌪 DALL-E 3 Turbo
- 📽 LUMA
- 🌑Luma Image
- 🎞 RunWay
- 👁️ RunWay Act Two
- 🦋 RunWay Aleph
- 🎑RunWay Frames
- 📼 Kling
- ⭕ Veo
- ⭕ Imagen 4
- 🔺 Flux
- 🎦 Hailuo MiniMax
- 🤡 Pika
- 🧿Seedream
- 👾Wan
- 🎯Grok Imagine
- 🎠 HappyHorse
❌ Prohibited Content:
1. Violations of Rules and Laws
- use of copyrighted materials, including copying or closely imitating well-known works, brands, or products, may be restricted
- fraud, spam, and user deception are not allowed
- misinformation and manipulative content may be blocked
- encouragement or promotion of illegal activities is not supported
2. Violence and Dangerous Content
- scenes of cruelty or violence, including violence toward animals, may be simplified, altered, or blocked
- content that threatens safety, health, or life is restricted
- weapons, their use, and related instructions may not be supported
3. NSFW and Sensitive Content
- pornography and explicit sexual scenes may be blocked or simplified
- excessively explicit, suggestive, or indecent content is restricted
- content related to drug use, controlled substances, or other harmful substances may not be supported
4. Discrimination and Offensive Content
- racism, sexism, hate speech, and discriminatory content are not allowed
- insulting, degrading, or humiliating content is restricted
- profanity or aggressive wording may trigger filters
5. Content Involving People
- use of public figures may be restricted, especially in sensitive, misleading, or inappropriate contexts
- scenes involving children or specific age references may be rejected
- any hints of exploitation, abuse, or unsafe situations are strictly prohibited
6. Scene and Phrasing Restrictions
- scenes that contradict commonly accepted norms may be rejected, altered, or simplified
- certain words and phrases may trigger filters, for example: “wet clothing,” “cleavage,” “breast,” “underwear,” “latex,” “nude,” “skin,” “girl,” “boy,” “waist,” “body,” etc.
7. Technical Limitations
- some restrictions are related to how the model interprets the scene, realism, or context
- use of styles associated with contemporary artists may be limited
❗ Additional Notes
- Some models may restrict generation when public figures, brands, copyrighted works, or recognizable products are mentioned.
- Restrictions may vary even within the same category of tools, as each model may interpret requests differently.
- Model behavior often depends on how the request is phrased, as well as on the overall context and intended meaning of the prompt.
Neural Networks with Minimal Filters
These tools have minimal restrictions and provide users with a high degree of freedom when working with content.
However, the user is fully responsible for the generated results. It is recommended to follow ethical principles, respect legal requirements, and keep moral standards in mind when creating and using generated content.
This category includes the following neural networks:
- 🎭 Face Swap
- 📸 Photo Maker
- 🧬 Stable Diffusion
- 🖌 inPaint
- ▪ Recraft
- ✂ Editor
- 🪁 Clarity Upscaler
- 🔍Magnific Upscaler
- 👠 Kling Try-On
- 🌈 Kling Kolors
- 🔺 FLUX
- 🎙 ElevenLabs Voice
- 👂 Audio to Text
- 🥁 Sound Generator
- 🌊 Video to Audio
- 👥 Voice Cloning
- 💠 Topaz AI
- ⚛️ Higgsfield
❌ Prohibited Content
- creation of materials that violate laws, regulations, or platform rules
- any form of child exploitation, abuse, or content that could be interpreted as sexualized content involving minors
- NSFW, explicit, or highly sensitive content
Additional Group of Tools
These tools do not directly fall into the categories of generation models with different levels of filtering, but they have their own limitations and usage rules.
Language models
💡 ChatGPT / GPTs / DeepSeek / Qwen / Claude / Gemini / Grok / Perplexity
Restrictions:
- creation, copying, or close imitation of copyrighted content
- illegal activities, fraud, and trade in prohibited goods
- malicious actions, including hacking, creating viruses, or bypassing security systems
- violence, suicide, self-harm, and dangerous activities
- pornography and explicit content
- discrimination, hate speech, and insults
Music Generation
- 🎸Suno
- 🎷ElevenLabs Music
Restrictions:
- use of lyrics or songs protected by copyright
- mentioning or imitating celebrities and public figures
- content related to the promotion of war or violence
Avatars and Speech
- 🗣D-ID Avatars
- ✴ Hedra
- 💜HeyGen
- 🫦Lipsync
Restrictions:
- use of profanity or offensive language
- creating avatars using the likeness of public figures
Important to Know
Users who repeatedly violate the terms of service may be blocked from the platform without prior notice.
It is also important to keep in mind that even when all recommendations are followed, a tool may still refuse to generate content due to internal limitations or the specific way the model operates.
Thank you for your attention and for taking the time to study this knowledge base on filters. We are confident that the knowledge you have gained will help you choose and use the right tools effectively according to your needs! 💛
SYNTX AI: Syntx AI
SYNTX Support: Syntx Support
YouTube channel “SYNTX”: Syntx YouTube
SYNTX.AI Academy: SYNTX Academy