Choose an animation workflow by the work that must remain editable after the first attractive render. For a short visual clip, Dreamina can supply reference-guided generated footage. For an explainer with exact labels, build the message from editable text and graphics, using generated footage where it helps. For a character that must perform repeatedly, decide whether recognizable rendered shots are enough or whether you need an actual puppet or 3D scene.
Those choices can produce similar-looking videos while leaving you with very different files. A rendered character waving does not give you an arm control. A beautiful animated diagram does not necessarily leave its numbers editable. The right starting question is: What will someone ask me to change next, and what file will let me make that change?
This guide follows three practical deliverables through production and handoff. The examples and prompts are proposed planning exercises, not additional software tests.
Name the deliverable before choosing the tool
Write a one-sentence brief that includes the output, the essential control and the likely revision. “Make a fun animation” is too open. “Deliver an eight-second looping welcome clip; the mascot must remain recognizable; the closing date will be added separately” gives you a usable production decision.
Original instructional diagram. It maps proposed deliverables and controls; it does not show software test results.
A template can fit the explainer route when its actual project exposes the text fields, scene timing and graphics you need. Previewing an animated template tells you what it looks like; opening and changing its project tells you whether it fits your revisions. A generative tool is useful for visual invention, while a template or puppet is useful when its existing structure matches the job. Neither label establishes quality by itself.
If someone requests both a finished MP4 and an editable character, put both in the brief. Delivering one does not implicitly deliver the other.
Route 1: Make a short clip when the footage is the product
Imagine a six-second welcome animation: an original round blue robot notices the viewer, raises one hand and holds a friendly pose. There is no spoken line and no text inside the generated scene. The delivery is a video, so reference-guided generation is a reasonable starting route.
The hard requirement is a readable greeting. Tiny idle movements can vary. If the brief instead requires the hand to follow an exact frame-by-frame path while a mechanical joint remains physically correct, move that requirement into an editable animation route before producing the full piece.
Prepare one image that can start the action
Define the robot's visible identity: round blue body, one amber circular eye, two short arms, two feet and no lettering. Place its resting hand away from the body so the later raise has a clear starting position. Leave space above the hand and keep the full silhouette visible.
Dreamina's Seedream 5.0 Pro instructions document opening AI Image, entering text or uploading an image, selecting Image 5.0 Pro, setting aspect ratio and resolution, then generating, reviewing and downloading. Use that path to prepare the starting artwork; review the actual design before treating it as a reusable reference.
A proposed image prompt:
One original round blue robot with one amber circular eye, two short arms and two feet, standing in a simple cream studio. Its right hand rests slightly away from its body, ready to rise. Show the whole robot, with clear empty space above its hand. Soft illustrated shading, readable silhouette, no writing, no extra characters.
Check the output against the design, not just the mood. If the robot has two eyes or a hand is hidden, fix the input before generating movement. The character reference preparation guide explains how to approve the master and choose the views a shot actually needs. A large sheet is unnecessary for this single frontal greeting if it resolves no additional uncertainty.
Generate a bounded action in Dreamina
The documented Seedance 2.5 workflow uses AI Video, the model selection and Omni reference inputs, with uploaded assets assigned through references in the prompt. Select the controls available in your account and identify the image's role. An image used as an identity reference is not automatically a guaranteed first frame.
For the robot, upload the accepted artwork and use a proposed motion instruction such as:
Use the uploaded robot image for its identity and visual style. Create one six-second frontal medium-wide shot in the same cream studio. During the first second the robot notices the viewer. From one to four seconds it raises its right hand into a friendly greeting. During the final two seconds it holds the raised hand with a small relaxed body movement. Keep the camera stationary and the whole robot inside the frame. Preserve one amber eye and the accepted blue body. No cut, lettering or new objects.
Set a supported duration that leaves enough material for the intended edit. Review the displayed settings and cost before submitting. A six-second instruction expresses the desired performance; it is not evidence that the gesture will reach the requested pose at precisely four seconds.
Approve the clip for its actual use
Watch the complete candidate at normal speed. Does the greeting read without explanation? Then inspect the opening, the hand's most complex moment and the ending. Look for a changing eye count, an arm merging into the body, a clipped hand and unintended camera movement.
A clip can pass with slightly different timing if the greeting remains clear and the final edit can use it. It fails when the action is ambiguous or the identity changes in a way that matters to the brief. Write the decision in terms of the requirement: “The hand reaches the greeting pose and holds long enough for the end card,” rather than “Looks cinematic.”
For a loop, compare the first and last visible states. A raised-hand ending will jump back to a lowered-hand opening. Either design a return to the resting pose, use a loop-friendly idle action, or deliver a non-looping greeting. Reversing the clip mechanically may reverse a blink or other movement awkwardly; inspect the repeated playback rather than assuming that a mirrored timeline solves it.
Finish and retain the source
Trim the approved section in an editor, add any required closing text on a separate layer and export the requested size and format. Open the saved file and check the entire result, including the final frame. Keep the original generated video alongside the edited version so a later trim does not require another generation.
This route succeeds when an approved piece of footage is the useful asset. If the next request is “raise the hand a little less while everything else remains exactly unchanged,” it may require a new candidate and another review. That revision cost belongs in the choice of workflow.
Route 2: Build an explainer around the information that must stay correct
Now imagine a 20-second explainer for a repair service. It needs three steps—book, bring, collect—and a price that may change next month. The critical assets are the sequence, the wording and the price. Generating all of them into the image would make a routine information update depend on rebuilding footage.
A useful proposed schedule is four seconds for the problem, five seconds each for the first two steps, four seconds for collection and two seconds for the call to action: 4 + 5 + 5 + 4 + 2 = 20 seconds. This is a starting edit plan, not a promise that every viewer can read the same amount of text in each segment. Preview it at the intended display size.
Separate the information from the illustration
Create a simple content table before making visuals:
Use Dreamina for an illustration or short contextual shot where that supplies useful visual character. Keep the price, contact details and step labels as text objects in the finishing project. A static icon animated with a simple position or scale change may communicate “book” more clearly than a generated hand operating a calendar.
Assemble a project you can actually revise
One documented desktop route is Premiere. Import the visual assets and place the segments on a timeline in the planned order. Create the labels with the Type tool; Adobe's title instructions describe editing their appearance in Properties. Name the price item so it can be found without searching through the picture.
Where a still needs gentle movement, select its clip and adjust Position or Scale through Motion in Effect Controls. Adobe documents those controls in its Motion effect guide. Use movement to reveal or emphasize information, not to keep every object in constant motion. A normal pan or zoom across a still does not create a new 3D viewpoint.
Record or import the approved narration, then adjust segment lengths around the actual spoken delivery. Keep the close readable after the last sentence rather than filling every available frame with speech. If there is no narration, the labels and visual transitions must carry the explanation themselves.
These are documented construction steps, not a claim that this hypothetical explainer has been produced. A template-based editor can serve the same purpose if the relevant text, timing and graphic elements remain independently editable in its saved project.
Perform the revision before calling the project reusable
Duplicate the working sequence and change the hypothetical price from 25 to 30. Does the new text fit? Does it cover the illustration? Does the voiceover still say the old price? This simple rehearsal tests the actual revision requirement without rebuilding the creative concept.
Then replace one step label with a longer approved phrase. If the design breaks, adjust the text area or typography in the master. Avoid solving the problem only in the exported video: that would leave the next update with the same defect.
The handoff should include the editor project, used media, current text and narration, plus the finished video. Reopen the project from its intended folder and check for missing assets. An MP4 is the delivery copy; the project and dependencies are the revision copy.
For export, Adobe's video and audio export guide documents File → Export → Media and H.264 for MP4 output. Confirm the actual frame size, duration and enabled audio/video before export, then open the resulting file. Matching a filename extension does not prove that all information survived the final render.
Route 3: Choose what “repeatable character” means in your project
A weekly mascot series can mean two different things. You might need the same recognizable character in a new short scene each week, accepting individually reviewed variations. Or you might need a character whose mouth, arm and expression controls can be reused for exact performances. The first is a reference-and-footage problem; the second is an animation-asset problem.
When approved references and generated shots are sufficient
Use an accepted design pack as the authority, create shot-specific images, generate bounded actions and retain the selected clips. Keep a record of which image entered each shot and which design version it represents. This makes changes traceable; it does not turn the references into a rig.
Consider a mascot that welcomes a visitor, points toward a notice and leaves. If broad gesture and recognition are sufficient, individual generated shots may meet the job. If the notice changes, add it as editable graphics. If the mascot's exact finger must touch a specific point on the notice at an exact moment, plan that interaction separately rather than hiding the requirement inside “keep consistent.”
Before producing a series, qualify the most demanding recurring action. A calm frontal greeting is a weak test for an episode that requires a full turn and an object handoff. Choose the shot that exposes the needed control, then decide whether the footage route remains appropriate.
When you need a reusable 2D puppet
A puppet starts with structured artwork, not a finished flat character image alone. Adobe's artwork preparation guidance explains how named and structured Photoshop or Illustrator layers support tags and behavior controls in Character Animator. Plan the parts needed for your performance: body, head, eyes, mouth states and movable arms, for example. A generated drawing can inform this artwork, but the required separation and hidden areas still need preparation.
Adobe's puppet documentation describes importing a PSD or AI file, adding the resulting puppet to a scene and exporting a self-contained .puppet asset. Its export excludes recorded takes and timeline-instance parameter changes. Keep the working project as well as the reusable puppet when the recorded performance must remain editable.
For a practical first qualification, build one neutral pose and one required gesture. Inspect the puppet in its scene, set the applicable behaviors, record a short performance and check that the arm or expression responds as intended. Adobe's performance tutorial demonstrates organized views, behaviors, recording and MP4 export through Adobe Media Encoder. Follow the controls for your prepared artwork; its hummingbird setup is an example, not a universal humanoid rig.
Accept the route only after saving, reopening and changing the intended control. If moving the arm reveals a blank shoulder or switching the mouth changes the face outline unexpectedly, repair the artwork or setup. A smooth exported preview cannot substitute for that control check.
When editable 3D is required
Choose a genuine 3D production path when the delivery needs geometry, a manipulable camera, materials or reusable spatial animation. Confirm that the supplied asset includes the controls your shots need and that the scene opens with its dependencies. A model without a usable animation setup is a different starting point from a ready-to-perform character.
The resulting scene and rendered video are separate deliverables. Maya's animation layers documentation, for example, describes keyframe curves and layers that can modify animation while retaining the original curves. A generated video with a sculpted appearance does not provide those controls. The same-story 2D-look versus 3D-look comparison explains how to separate a visual treatment choice from a source-asset requirement.
Use one revision request to make the final decision
Before committing to the full project, write the change most likely to arrive after approval and identify its smallest editable unit.
Budget around approved deliverables and likely revisions, not only the first generation. Include time for input preparation, rejected candidates, artwork setup, editing and export checks. A route with more initial preparation can be sensible for repeated exact changes; a generated clip can be sensible when a single reviewed visual result is enough. Neither is automatically cheaper for every project.
If the route works but the story remains vague, solve the action allocation before producing additional candidates. The worked storyboard and shot-prompt guide turns one new story into timed cards with explicit start and end states.
Start with the smallest complete sample that represents the real job: one finished short clip, one fully updateable explainer segment, or one saved character asset performing a required gesture. Use Dreamina where generated artwork and footage serve that deliverable, and retain the editable project wherever the next revision requires direct control.