Video generation in Layer uses third-party model providers including Kling (Tencent), Veo (Google), Sora, Seedance, and others. Most issues fall into one of a handful of categories. This article covers how to identify what is happening and what to do about it.
Common issues at a glance
Issue | Most likely cause |
Generation fails with a generic error | Model provider instability or a session state issue |
Generation runs forever without completing | Provider slowness, especially with newer Kling models; input video too short for model requirements |
Safety or NSFW rejection on non-explicit content | AI content safety false positive; model-specific sensitivity |
Output video crops or misframes the reference image | Aspect ratio mismatch between input image and selected output ratio |
First or last frame option not available on the selected model | Not all models support both reference frames; support varies by Kling version |
Prompt button not activating in the video section | Video edit models require Prompt Edit, not the standard Video Forge |
Reframe producing incorrect or cropped output | Resolution mismatch or content safety trigger during reframing |
Generation fails with an error
If a video generation returns an error immediately or partway through, the most likely cause is provider-side instability. Check whether the issue is isolated to one specific model or affecting all video models. If only one model is affected, this almost always points to the provider rather than Layer.
What to try:
Retry the generation once. Some errors are transient and a second attempt on the same model will succeed.
Switch to an alternative model. If Kling 3 is failing, try Veo 3.1 or Kling 2.6 Turbo. If Veo is failing, try Kling or Seedance Pro.
Check the Layer Status Page for any active incidents.
If the error persists across multiple models, start a fresh session from the Layer home page and try again.
Note that failed video generations do not deduct Creative Units. You are only charged when a generation completes successfully.
Generation is very slow or appears stuck
Video generations take longer than image generations by nature, but some models can be significantly slower during periods of high demand or when the model provider is experiencing optimisation issues. This has been most consistently seen with Kling 3 models.
What to try:
For Kling 3 slowness, Kling o3 Pro tends to complete faster than Kling v3 Standard. Try switching to that specific variant.
Veo 3.1 is the recommended alternative when Kling is slow, though it does not support square aspect ratios.
If a generation has been running for an unusually long time and appears frozen, refresh the page. The generation may have completed in the background and will show in your activity feed.
Check that your input video meets the minimum length requirement for the model. Kling O1, for example, requires input videos between 3 and 10 seconds. Videos shorter than 1 second will cause a failure rather than a slow generation.
Safety or NSFW rejection on non-explicit content
Video models use AI-based content safety checks that can produce false positives on completely non-explicit content. Gaming assets, character animations, vehicle transformations, and hand movements have all been flagged incorrectly in practice. This is a behaviour of the model provider's safety system and is not something Layer can override directly.
What to try:
Retry the generation with the same prompt and model. The safety check runs fresh each time and the same content may pass on a subsequent attempt.
Simplify the prompt. Very detailed or descriptive prompts occasionally trigger safety systems. Try a shorter, more neutral version of the same instruction.
Switch to a different model, as each provider has its own content safety policy. If Veo is rejecting your content, Kling or Sora 2 may accept it, and vice versa.
Note that Veo models have a stricter content policy around character generation. If you are trying to animate characters and Veo keeps rejecting the prompt, Kling 2.5 Turbo or Sora 2 are more permissive alternatives for this use case.
Output video crops the reference image
If the generated video only shows a portion of your reference image or the subject appears cropped, this is almost always caused by a mismatch between your reference image's aspect ratio and the aspect ratio selected for the video output.
What to try:
Match your reference image's dimensions to the output aspect ratio you have selected. If you are generating a 16:9 video, upload a 16:9 reference image. If you are generating 9:16, use a portrait reference image.
You can adjust the reference image framing inside Layer's reference editor before generating.
If you need to change the aspect ratio of an already-generated video, use the Reframe feature. Open the video, select your target ratio, and write a prompt describing how the frame should be expanded — for example, "expand the sides to fit landscape format" or "extend the top and bottom to fit portrait."
Note that Veo 3.1 does not support square (1:1) aspect ratio. If you need square output, use a Kling model instead.
First frame and last frame reference support
Not all video models support both first and last frame references, you can check if a model supports this on Layer's Models page.
If a model you need for first/last frame support has been removed from your workspace, a workspace Admin can re-add it from the model settings. If you are not seeing a model that should be available, check with your workspace Admin before contacting support.
Editing an existing video with Prompt Edit
If you want to modify an existing video rather than generate a new one from scratch, you need to use Prompt Edit for Video rather than the standard Video Forge. Using a video editing model inside the Video Forge will result in errors or the Forge button not activating.
To edit a video with Prompt Edit, navigate to the Prompt Edit section from the Layer home page, select Video as the modality, upload your source video, and enter your editing prompt. For Kling 3 specifically, you can reference multiple images in Prompt Edit using @Image1, @Image2, and so on for more controlled results.
Reframe issues
The Reframe tool changes the aspect ratio or framing of an existing video. If Reframe is producing unexpected output or failing with a content safety rejection:
Check that your source video is at least 1080p. Lower resolution sources can produce poor Reframe results due to limited pixel data for the model to work with.
If Reframe is returning a safety rejection on content that is clearly not explicit, retry once and then try a different model. Ray 2 and Wan 2.2 are available alternatives for reframing depending on your workspace configuration.
Write a specific reframe prompt describing the desired output, for example "extend the left and right sides of the video to fit 16:9 landscape format." Vague or empty prompts can produce inconsistent results.
Frequently Asked Questions
My video failed partway through. Were CUs charged?
No. Creative Units are only deducted when a generation completes successfully. A failed video generation does not charge your balance. You can confirm this by checking Workspace > Usage.
How long should video generation normally take?
Generation time varies significantly by model and duration. Shorter videos on faster models can complete in 2 to 5 minutes. Longer videos on premium models like Kling 3 can take 10 to 15 minutes or more during periods of high demand. If a generation exceeds 30 minutes without completing, try refreshing the page to check your activity feed, then contact support with your session link if it is still not there.
Veo keeps rejecting my character generation prompts. What should I do?
Veo models apply stricter safety rules around character content. For character animations, Kling 2.5 Turbo or Sora 2 are generally more permissive and better suited for this use case.
Can I use my trained LoRA model with video generation?
No. Custom trained LoRA models apply to image generation only. Video generation models do not currently support LoRA styles.
The output video looks nothing like my reference image. What can I do?
First check that your reference image matches the output aspect ratio you selected. If the ratios match and the output still drifts significantly from the reference, try a more descriptive prompt that includes specific visual details from the image, or try a different model. Kling models with first/last frame support tend to give more reference-faithful results than text-only prompts.
Can I generate a video longer than 10 seconds?
Some models support up to 15 seconds per generation, including Kling 3 which supports 3 to 15 second videos. For longer outputs, you can generate sequential clips and stitch them together using the Workflows feature, using the Select Frame tool to extract a keyframe from one clip as the first frame of the next.
Still experiencing issues with video generation? Contact support with your session link and the specific model and prompt you were using and we will investigate.