Every tool starts at zero and earns points for what it does and what it costs. Here is exactly what earns them.
The short version
We check each tool on 5 things a buyer cares about. Together they are worth 100 points.
The final score is those points out of 10. A tool with 84 points scores 8.4.
Everything comes from the tool's own website. We don't take payment for points.
For videoFor imagesPricingWorkflow and teamsPrivacy and trust
What earns points
1. For video
30 of 100 pointsHow good a place it is to make video: how many model families it offers from other companies (six versions of Kling count as one Kling), whether the current ones are there (Kling 3, Veo 3.1, Seedance 2), and lip sync, extending a clip and setting the first and last frame.
Video model families offeredup to 3
| None | 0 |
| 1 to 2 video model families | +0.5 |
| 3 video model families | +1 |
| 4 to 5 video model families | +2 |
| 6 to 8 video model families | +2.5 |
| 9 video model families or more | +3 |
Extends a clip past its endup to 1
You set the first and last frameup to 1
Each section adds up to 10, then counts for 30 of the 100.
2. For images
30 of 100 pointsThe same for images: how many image model families, whether Nano Banana Pro, GPT Image 2 and Flux 2 are there, and an upscaler, image editing and training a model on your own pictures.
Image model families offeredup to 3
| None | 0 |
| 1 to 2 image model families | +0.5 |
| 3 image model families | +1 |
| 4 to 5 image model families | +2 |
| 6 to 8 image model families | +2.5 |
| 9 image model families or more | +3 |
Offers Nano Banana Proup to 1.5
Offers GPT Image 2up to 1.5
Edits images (inpaint, erase, relight)up to 1
Trains a model or LoRA on your imagesup to 1
Each section adds up to 10, then counts for 30 of the 100.
3. Pricing
15 of 100 pointsWhat the cheapest paid plan costs each month, what any unlimited generation covers (all-year unlimited on other companies' video models scores highest, then image models, then relaxed mode on the platform's own models), whether there is a free plan, and whether you may use the results commercially. Most platforms show what one generation costs only after sign-in, so the plans themselves are compared.
Cheapest paid plan, monthlyup to 4
| Up to $10 | +4 |
| Up to $20 | +3.5 |
| Up to $30 | +2.5 |
| Up to $50 | +1.5 |
| Over $50 | +0.5 |
What unlimited generation coversup to 3
| third_party_video | +3 |
| third_party_image | +2.5 |
| own_models | +1.5 |
| No | 0 |
Commercial use allowedup to 1
Each section adds up to 10, then counts for 15 of the 100.
4. Workflow and teams
12 of 100 pointsWorking at scale: a canvas or node editor that chains generations, an AI agent, an API, a shared team workspace and where it runs.
Canvas or node workflowsup to 2.5
An AI agent that runs multi-step jobsup to 2
Shared team workspaceup to 2
Where it runsup to 1.5
| 1 platform | +0.5 |
| 2 platforms | +1 |
| 3 platforms or more | +1.5 |
Each section adds up to 10, then counts for 12 of the 100.
5. Privacy and trust
13 of 100 pointsDoes it train its AI on what you upload, does it hold SOC 2 and state GDPR, is there single sign-on, and does the cheapest paid plan remove the watermark?
Does not train AI on your uploadsup to 4
| Yes | +4 |
| Only if you opt out | +2 |
| No | 0 |
The cheapest paid plan exports without a watermarkup to 1.5
Each section adds up to 10, then counts for 13 of the 100.
A real example: Magnific
Here is how Magnific got its 9.4.
| Section | Points |
|---|
| For video | 28.5 of 30 |
| For images | 30 of 30 |
| Pricing | 11.3 of 15 |
| Workflow and teams | 12 of 12 |
| Privacy and trust | 12 of 13 |
| Total | 93.8 of 100 = 9.4 out of 10 |
See every fact behind Magnific's score.
When a tool doesn't say
If a tool's website doesn't answer a yes-or-no question, it gets half the points for that item. Not zero, because a missing answer isn't the same as a no, and not full marks, because silence shouldn't beat a vendor who answers honestly. For a number, like a price, it gets the typical value of the tools that do publish one. Every item scored this way is marked on the tool's scorecard.
If a tool leaves out more than a third of what we check, we don't show a score for it at all. A number built mostly on guesses wouldn't be fair to you or to the tool. Right now 13 of the 14 tools we checked have enough information to be scored.
What the score doesn't tell you
We score what a tool offers and what it costs, as its own website describes it. That's not a hands-on test. The score can't tell you whether a tool is easy to use, whether its data is accurate, or whether support answers quickly. The review covers that.
See the Multi-model AI platforms ranked or read how we review tools in general.