Let's put it to the test!
Choosing an AI tool today is about more than selecting the one with the most impressive output. For professional use, you also need to know what happens to your data, whether you are allowed to use the output commercially, and whether you can explain to clients why you chose that particular tool.
Below you will find an overview of commonly used AI tools. We do not evaluate them based on creative quality, but on trust, transparency and professional suitability. Stay critical, as the privacy policies, training practices, commercial rights and data retention periods of these tools change regularly!
How to read this overview
We use three indicators:
✓ = Clear or favourable
~ = Depends on the subscription, settings or usage
? = Insufficiently clear
The tools we have tested:
ChatGPT
ChatGPT is suitable for professional use when you work through ChatGPT Business, Enterprise, Edu or the API. OpenAI states that input and output from business products are not used by default to train models. Standard consumer accounts do offer settings to manage data usage.
| Check | Assessment |
|---|---|
| Clear provider | ✓ |
| Clear what happens to your input | ~ |
| Ability to disable training on your data | ✓ |
| Commercial use | ~ |
| Copyright clarity | ~ |
| Faces, voices and deepfakes | ~ |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | Limited, depending on the output |
| Data location sufficiently clear | ~ |
| Suitable for professional use | ✓ With business accounts |
Advice: For client work or confidential information, preferably use a business account.
Google Gemini and Veo
With Google, it is important to distinguish between personal use and use within Google Workspace. Google states that chats and uploaded files in Gemini within Workspace are not reviewed by human reviewers or used to train generative AI models outside your domain without permission. Different privacy conditions apply to personal use.
| Check | Assessment |
| Clear provider | ✓ |
| Clear what happens to your input | ~ |
| Ability to disable training on your data | ~ |
| Commercial use | ~ |
| Copyright clarity | ~ |
| Faces, voices and deepfakes | ~ |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | ~ |
| Data location sufficiently clear | ~ |
| Suitable for professional use | ✓ Within Workspace, more caution with personal accounts |
Advice: For professional work, preferably use Gemini or Veo within a business Google Workspace environment.
Claude
Claude makes a clear distinction between consumer accounts and commercial services. Anthropic states that users of Free, Pro and Max can choose whether their data may be used to improve Claude. This does not apply to commercial services such as Claude for Work, Claude for Education, Claude Gov and API usage.
| Check | Assessment |
| Clear provider | ✓ |
| Clear what happens to your input | ~ |
| Ability to disable training on your data | ✓ |
| Commercial use | ~ |
| Copyright clarity | ~ |
| Faces, voices and deepfakes | Limited relevance |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | Limited |
| Data location sufficiently clear | ~ |
| Suitable for professional use | ✓ With commercial accounts |
Advice: For professional work, use Claude for Work or the API rather than a personal account.
Microsoft Copilot
Microsoft 365 Copilot is particularly interesting for organisations already working within Microsoft 365. Microsoft states that prompts and responses within Microsoft 365 Copilot are protected under enterprise data protection and are not used to train foundation models.
| Check | Assessment |
| Clear provider | ✓ |
| Clear what happens to your input | ✓ |
| Ability to disable training on your data | ✓ With Microsoft 365 Copilot |
| Commercial use | ~ |
| Copyright clarity | ~ |
| Faces, voices and deepfakes | Limited relevance |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | Limited |
| Data location sufficiently clear | ~ |
| Suitable for professional use | ✓ With Microsoft 365 Copilot |
Advice: Distinguish between the free version of Copilot and Microsoft 365 Copilot with enterprise protection.
Adobe Firefly
Adobe explicitly positions Firefly as an AI tool for professional and commercial use. Adobe refers to a responsible training approach and Content Credentials, which make it possible to include information about the provenance of AI-generated content.
| Check | Assessment |
|---|---|
| Clear provider | ✓ |
| Clear what happens to your input | ✓ |
| Ability to disable training on your data | ~ |
| Commercial use | ✓ |
| Copyright clarity | ✓ |
| Faces, voices and deepfakes | ~ |
| AI use disclosure | ✓ Via Content Credentials |
| Provenance or watermark | ✓ |
| Data location sufficiently clear | ~ |
| Suitable for professional use | ✓ |
Advice: Firefly is one of the safer choices for commercial creative workflows, especially when provenance is important.
Midjourney
Midjourney states that paying users are allowed to use images and videos commercially, with certain exceptions. The tool is strong for visual creation, but less transparent than some business platforms regarding data, provenance and copyright certainty.
| Check | Assessment |
| Clear provider | ✓ |
| Clear what happens to your input | ~ |
| Ability to disable training on your data | ~ |
| Commercial use | ✓ With a paid account |
| Copyright clarity | ~ |
| Faces, voices and deepfakes | ~ |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | Limited |
| Data location sufficiently clear | ~ |
| Suitable for professional use | ~ |
Advice: Suitable for concepts, moodboards and visual exploration. Be particularly cautious when using recognisable styles, people and brands in final work for client campaigns.
Runway
Runway is a professional platform for AI video. Runway states that commercial use of its output is not restricted, but also that input and output may be used to train and improve its models and services.
| Check | Assessment |
| Clear provider | ✓ |
| Clear what happens to your input | ✓ |
| Ability to disable training on your data | ~ |
| Commercial use | ✓ |
| Copyright clarity | ~ |
| Faces, voices and deepfakes | ~ |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | ~ |
| Data location sufficiently clear | ~ |
| Suitable for professional use | ~ |
Advice: Strong for creative production, but do not upload confidential client material without the appropriate account terms.
Freepik
Freepik is more of a platform than a single AI model. It combines various creative tools and models. Freepik states that images and voices uploaded by users are not used to train its own AI models or those of third parties. For AI-generated content, Freepik also allows commercial use, depending on the subscription plan.
| Check | Assessment |
| Clear provider | ✓ |
| Clear what happens to your input | ✓ |
| Ability to disable training on your data | ✓ For images and voices |
| Commercial use | ~ |
| Copyright clarity | ~ |
| Faces, voices and deepfakes | ~ |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | ~ |
| Data location sufficiently clear | ~ |
| Suitable for professional use | ~ |
Advice: Suitable for creative production, but check the applicable terms for each tool or model available within Freepik.
Flora
Flora is a creative environment that brings together multiple AI tools. As a result, you should not only evaluate Flora itself, but also the underlying models. Flora states that uploaded files, prompts and metadata are not used to train models, including by model partners.
| Check | Assessment |
|---|---|
| Clear provider | ✓ |
| Clear what happens to your input | ✓ |
| Ability to disable training on your data | ✓ |
| Commercial use | ~ |
| Copyright clarity | ~ |
| Faces, voices and deepfakes | Depends on the underlying tool |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | Depends on the underlying tool |
| Data location sufficiently clear | ~ |
| Suitable for professional use | ~ |
Advice: An interesting creative hub, but check which underlying AI tool you are using for each project.
Stable Diffusion
Stable Diffusion is not a standard web tool, but a family of models. You can run them locally or use them through third-party platforms. This makes them attractive from a privacy perspective, as running them locally gives you full control over your data. However, the licence differs depending on the model version and provider.
| Check | Assessment |
| Clear provider | ✓ |
| Clear what happens to your input | ✓ When used locally, ~ via platforms |
| Ability to disable training on your data | ✓ When used locally |
| Commercial use | ~ Depends on the licence |
| Copyright clarity | ~ |
| Faces, voices and deepfakes | ~ |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | User’s responsibility |
| Data location sufficiently clear | ✓ When used locally |
| Suitable for professional use | ~ |
Advice: A strong option for organisations that want full technical control. Always verify the licence of the specific model.
Flux
Like Stable Diffusion, Flux is a family of models. Some versions are more freely usable than others. For example, the FLUX.1 dev licence is not intended for commercial or production use without the appropriate permission. Using Flux through partners such as Adobe, Freepik or other platforms may be subject to different terms.
| Check | Assessment |
| Clear provider | ✓ |
| Clear what happens to your input | Depends on usage |
| Ability to disable training on your data | ✓ When used locally |
| Commercial use | ~ Depends on the model and licence |
| Copyright clarity | ~ |
| Faces, voices and deepfakes | ~ |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | User’s responsibility |
| Data location sufficiently clear | ✓ When used locally |
| Suitable for professional use | ~ |
Advice: Use Flux professionally only if you are certain which version you are using and which licence applies to it.
LTX
LTX is a video model and platform developed by Lightricks. Its licensing structure differs by version. For some open-weight versions, commercial use is not permitted for commercial entities without a separate licence. The LTX website also mentions options for on-premise deployment and commercial licences.
| Check | Assessment |
| Clear provider | ✓ |
| Clear what happens to your input | Depends on usage |
| Ability to disable training on your data | ✓ When used locally or on-premise |
| Commercial use | ~ Depends on the licence |
| Copyright clarity | ~ |
| Faces, voices and deepfakes | ~ |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | User’s responsibility |
| Data location sufficiently clear | ✓ When used locally |
| Suitable for professional use | ~ |
Advice: Verify the exact LTX version and its licence before using its output in client work.
Wan
Wan is associated with Alibaba and exists both as a model and as a platform. The Wan terms apply when using it through wan.video. Open model versions are also available through other platforms. As a result, the assessment depends heavily on where and how you use Wan.
| Check | Assessment |
| Clear provider | ✓ |
| Clear what happens to your input | ~ |
| Ability to disable training on your data | ? |
| Commercial use | ~ |
| Copyright clarity | ~ |
| Faces, voices and deepfakes | ~ |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | User’s responsibility |
| Data location sufficiently clear | ~ |
| Suitable for professional use | ~ |
Advice: Use Wan for professional work only after carefully reviewing the platform terms or the applicable model licence.
Kling
For Kling, we found platform terms and references to a privacy policy, but not enough clear official information to reliably complete all ten checks. Therefore, a cautious assessment is appropriate.
| Check | Assessment |
| Clear provider | ~ |
| Clear what happens to your input | ? |
| Ability to disable training on your data | ? |
| Commercial use | ~ |
| Copyright clarity | ? |
| Faces, voices and deepfakes | ~ |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | ? |
| Data location sufficiently clear | ? |
| Suitable for professional use | Only with caution |
Advice: For now, use Kling primarily for experimentation and concept development. Be cautious with client data, recognisable people and commercial publication.
Seedance and Seedream
Seedance and Seedream are models developed by ByteDance. Official information is available about the models themselves, as well as a privacy policy from the ByteDance Seed Team. At the same time, there are public concerns and reports regarding copyright, well-known characters and likeness issues associated with Seedance. Therefore, extra caution is required.
| Check | Assessment |
|---|---|
| Clear provider | ✓ |
| Clear what happens to your input | ~ |
| Ability to disable training on your data | ? |
| Commercial use | ~ |
| Copyright clarity | ? |
| Faces, voices and deepfakes | High risk |
| AI use disclosure | User’s responsibility |
| Provenance or watermark | ? |
| Data location sufficiently clear | ~ |
| Suitable for professional use | Only with strong controls |
Advice: Do not use Seedance or Seedream indiscriminately for commercial campaigns featuring recognisable people, well-known styles, brands or characters.
Which tools are the safest for professional use?
Based on the information currently available, the following tools appear to provide the greatest clarity for professional use:
- Adobe Firefly
- Microsoft 365 Copilot
- ChatGPT Business, Enterprise or API
- Claude for Work or API
- Google Gemini within Google Workspace
- Flora, provided that you take the underlying models into account
These tools offer the strongest guarantees regarding data handling, business terms or professional use.
Which tools require extra caution?
The following tools are creatively interesting but require more careful consideration:
- Midjourney
- Runway
- Freepik
- Stable Diffusion
- Flux
- LTX
- Wan
- Kling
- Seedance
- Seedream
This does not mean that you should not use them. It does mean that, for each project, you should verify whether client data, personal data, commercial rights and copyright are adequately protected.
Quick decision guide
Are you working with confidential client information?
Choose a business environment such as ChatGPT Business, Microsoft 365 Copilot, Claude for Work, Google Workspace or a locally hosted model.
Are you creating commercial images or videos for clients?
Always verify the licence and avoid recognisable people, brands, characters or styles without permission.
Are you mainly experimenting?
Then you can test more broadly, but do not use sensitive data.
Do you want maximum control?
Use locally hosted models, but always verify the model licence and provide your own labelling, storage policy and documentation.
Final thoughts
An AI tool is not safe simply because its output looks good. An AI tool is only truly suitable for professional use when you understand what happens to your input, what you are allowed to do with the output, and how you can explain your choice to a client, colleague or audience.
So let the demo inspire you, but make your decision only after completing the safety check.
Sources: OpenAI, Google, Anthropic, Microsoft, Midjourney, Runway, Freepik, Adobe, Stability AI, Black Forest Labs, Lightricks, Flora, ByteDance and Wan.