so many errors.

Subject: Serious Issues With Image-to-Video Generation in Firefly – Reference/First & Last Frame Errors

Hello Adobe Firefly Team,

I would like to report a frustrating and recurring issue I have been experiencing with Firefly's AI video generation tools.

I am a paying Firefly user and I have been trying to create a very simple cinematic image-to-video sequence using an image that I generated within Firefly itself. However, I have repeatedly encountered errors and limitations that prevent me from completing the workflow.

My intended workflow is the following:

I created a cinematic, fictional battle aftermath scene featuring an adult male character sitting in a natural environment. The character is tired and battle-worn, with mud and minor non-graphic blood marks on his clothing and face. He is sitting near a rock with a rifle resting beside him.

I want to create a short cinematic video in which:

• The character initially sits upright and slightly away from the rock.
• The camera slowly moves from approximately a 60-degree angle toward a 45-degree angle.
• The character remains mostly still during the camera movement.
• During the final few seconds, he slowly leans backward and rests against the rock.
• Once he is resting against the rock, he takes one deep, slow breath.
• The environment, character, face, clothing, rifle, injuries, lighting and composition should remain consistent throughout the shot.

I tried to accomplish this using Firefly's video generation tools, particularly the Kling 3.0 model.

The problems I encountered are:

1. Reference image problems

Some models appear to reject reference images containing a human face, with errors such as “reference cannot contain face.”

This is particularly confusing because I am trying to use an image of the same adult character throughout the video, and maintaining facial identity is one of the main reasons I need an image-to-video workflow in the first place.

2. First Frame / Last Frame problem

I discovered that Kling 3.0 provides a workflow for using a First Frame and a Last Frame.

This is exactly what I need.

I created two images showing the same character and the same environment:

• First frame: the character is sitting upright and has not yet leaned against the rock.
• Last frame: the character is fully resting against the rock.

However, when I attempt to upload the second image into the Last Frame field, Firefly repeatedly gives me a generic error such as:

“Something went wrong.”

There is no useful explanation of what caused the failure.

The first image can be uploaded successfully, but the second image repeatedly fails in the Last Frame field.

3. The images themselves are generated by Firefly

This is especially frustrating because these are not random external images. They are images generated specifically for this project using Firefly.

Therefore, I would expect Firefly's own video models to handle them reliably, especially when using the officially supported First Frame / Last Frame workflow.

4. I also tried adjusting the image dimensions

I suspected that the problem might be related to aspect ratio or resolution, so I recreated the images in a vertical format suitable for Instagram (9:16).

Even after doing this, the Last Frame upload problem persisted.

5. Different models produce different unexplained restrictions

I have tried multiple models available inside Firefly, including:

• Kling 3.0
• Kling 3.0 Omni
• Veo 3.1
• Veo 3.1 Fast
• Seedance 2.0
• Seedance 2.0 Fast
• Runway Gen-4.5
• Ray models

Instead of being able to determine which model is appropriate for the task, I often encounter different restrictions, errors, or reference-image limitations depending on the model.

This makes the workflow extremely difficult to understand.

I am not trying to generate anything explicit, graphic, or unsafe. This is a fictional cinematic scene intended for creative/video production. The character is an adult, and the injuries are deliberately non-graphic and cinematic.

What makes the situation particularly frustrating is that I am paying for Firefly specifically to have access to these creative tools, yet I have been unable to successfully complete even this relatively straightforward image-to-video sequence.

I would really appreciate it if the Firefly team could investigate:

• Why the Last Frame upload produces a generic “Something went wrong” error.
• Whether there is currently a technical issue with Kling 3.0 and Last Frame uploads.
• Whether there are undocumented restrictions on faces in First Frame / Last Frame workflows.
• What exact image dimensions, format, resolution and aspect ratio are required.
• Whether First Frame + Last Frame is fully supported for Kling 3.0 inside Firefly.
• Why an image generated by Firefly itself may be rejected or fail when used as a video reference.
• Why the error messages do not provide enough information to diagnose the problem.

Please provide a clear explanation of the correct workflow for creating an image-to-video sequence with a consistent human character using First Frame and Last Frame in Firefly.

At the moment, the experience feels unnecessarily difficult and unpredictable. I would really like to use Firefly for my video projects, but these problems are preventing me from doing so.

Thank you, and I hope this can be investigated.

Best regards,
A frustrated but genuinely interested Firefly user