Adobe publishes exactly one prompt formula for Firefly video, and it fits in a line: Shot Type Description + Character + Action + Location + Aesthetic.
Everything else on Adobe’s prompt page elaborates those five slots. This guide keeps the order, adds the official camera vocabulary that sits beside the prompt field, and marks where the documentation stops short.
The five-slot structure Adobe publishes
The formula comes from Adobe’s prompt guidance, and the same structure appears in Adobe’s 2024 announcement post for the beta, so it is not a one-off suggestion.
| Slot | The question Adobe asks | Adobe’s own example |
|---|---|---|
| Shot Type Description | What is the camera perspective? How is it moving? | a close-up shot with a slow zoom-in. |
| Character Description | Who is the character? What do they look like? What are they wearing? What is their emotion? | a large polar bear with bright white fur looking pensive. |
| Action | What is the character doing in the scene? | the polar bear is walking softly toward a hole it opened in the ice. |
| Location | Where is the character? What is the weather? What is the terrain? | the location is barren and snowy; gray clouds move slowly in the distance. |
| Aesthetic | What type of shot is it? What is the ambience? What is the depth of field? | cinematic, 35mm film, highly detailed, shallow depth of field, bokeh. |
Adobe’s framing sentence is blunt about density: use as many words as possible to be specific about lighting, cinematography, color grade, mood, and aesthetic style.
What belongs in each slot
Shot Type Description
Name the framing, then the movement. The official example pairs a size with a motion, and that pairing is the point of the slot.
Character Description
Who, what they look like, what they wear, what they feel. Adobe’s polar bear answers four questions in a dozen words, which is the density the slot asks for.
Action
One concrete verb plus pace. The official example uses walking softly but confidently and adds a destination, so the model has somewhere to move rather than a mood to hold.
Location
Weather and terrain, not a place name. The official example gives barren and snowy, then animates the sky.
Aesthetic
The technical finish: stock, detail level, depth of field. Adobe’s example is a comma-separated stack, and that stack is the pattern its own blog post repeats.
Examples: two official prompts, slot by slot
Both prompts below are Adobe’s, published on its prompt page under headings that name the shot. Neither is a template to fill in.
Example 1: the close-up portrait
Cinematic closeup and detailed portrait of a golden retriever dog in a field of sunflowers at golden hour. The lighting is cinematic, gorgeous, and soft, with beautiful, strong backlight and lens flare. The color grade is warm, sun-drenched, and sunlit. The dog is extremely realistic with detailed fur texture. The movement is subtle and soft. The camera doesn't move. There is heavy film grain and textures.Shot size: Close up shot
Camera angle: None
Motion: StaticAdobe files this under Close-up and explains the intent: a close-up is often used to highlight specific details or expressions. The prompt writes the framing into the sentence instead of leaving it to the dropdown, which is why the shot survives a change of resolution.
Example 2: the night street pan
A lonely street with cobbled streets and Victorian street lamps late at night bathes the scene in the warm yellow glow of the street lamps. A beautiful, vivid full moon adorns the city, and the camera pans slowly from right to left, revealing an abandoned Mansion in the distance.Shot size: None
Camera angle: Eye level shot
Motion: Move leftAdobe files this under Move, where the stated purpose is to give a sense of motion or focus to a subject. The prompt writes the pan into the sentence, and the Motion dropdown offers the matching value without the model having to infer it.
Camera controls that sit beside the prompt
Camera vocabulary appears on more than one Adobe surface. In the web app it is three dropdowns, and the same three ideas return as fields in the Firefly video API.
| Group | Options | Default |
|---|---|---|
| Shot size | Extreme close up, Close up shot, Medium shot, Long shot, Extreme long shot | None |
| Camera angle | Aerial shot, Eye level shot, High angle shot, Low angle shot, Top down shot | None |
| Motion | Zoom in, Zoom out, Move left, Move right, Tilt up, Tilt down, Static, Handheld | None |
None is not a placeholder in that table. Adobe states that by default there is no prescribed motion on the camera unless you select a Motion option or describe the camera motion in the prompt. The dropdown is an override, not a requirement.
When the controls are taken away from you
Two moves disable camera settings silently, and both are worth knowing before you blame the prompt.
Upload a motion reference video and Shot size, Camera angle and Motion get disabled, because the reference now supplies the movement. Add a first or last frame and the casualty list grows: Reference under Composition, Motion reference, Shot size, Camera angle and Style are all automatically disabled. Style works in the other direction, since selecting one disables the ability to set a start or end frame.
That answers most gray-out questions. The prompt slot keeps working in every one of those cases, which is why writing the shot type into the sentence pays off.
Length and subject limits
Adobe publishes two numbers and one caution. There are no hard limits on how long or short a prompt should be, but Firefly has a maximum word limit of 1800 words, and long prompts don’t always yield better results. On subjects, more than four can often confuse Firefly.
Neither number is a target. The 1800 is a ceiling, and the subject count is the one that changes how a prompt reads: five named characters in one line is a different risk from five nouns in an aesthetic stack.
Iterate instead of rewriting
Adobe’s advice is to start basic and add detail per run. Its page demonstrates that with one subject across four prompts.
Full scene, eye level, wide shot, a giant mech in the street, high quality, high details.
Full scene, eye level, wide shot, a giant mech, yellow and orange armor, in desolate streets, wires, LEDs, cybernetic parts, high quality, high details.
Full scene, eye level, wide shot, a giant black and yellow mech, accents of orange neon LEDs, marching, firing lasers, scanning, in the street of a destroyed city, rubble, fires, decayed buildings, desolate, ominous, high quality, great details.
Cinematic action scene, a group of giant mechs is invading the city, they are menacing, giant black and yellow mechs, yellow and orange matte armor, a dystopian future, in the street of a destroyed city, rubble, fires, decayed buildings, desolate, ominous, high quality, high details, volumetric lighting.The last prompt drops eye level and wide shot for cinematic action scene. That is the slot doing its job: framing moves from literal instruction to genre, and the rest of the prompt absorbs the difference. Once a generation lands, the Seed option makes the same seed, prompt and control settings reproduce similar clips, which is how the step-by-step guide treats a keeper. The credit rates cover what those repeated runs cost, and the model comparison explains why one prompt behaves differently once you switch the Model dropdown.
Frequently asked questions
Does the structure change with the model? The formula is Adobe’s guidance for its own model. Adobe states separately that the settings available for video generation may differ depending on the video model you select.
Do I need the camera dropdowns if I describe the shot in words? No. Adobe says there is no prescribed motion by default unless you pick a Motion option or describe the movement in the prompt.
What is the longest prompt I can write? Firefly has a maximum word limit of 1800 words, and Adobe notes that long prompts don’t always yield better results.
How many subjects can one prompt carry? Fewer than five. Adobe warns that more than four subjects can often confuse Firefly.