This page goes failure by failure — what it looks like, why it happens, what to change first — then gives the order to try those fixes in, so you are not spending credits on guesses.
Face warp and hands that melt
A face slides into a different person, or a hand folds into too many fingers. It is the most common complaint and the most fixable, because the cause is almost always the reference: a face that is small, dark, turned away or partly covered gives the model too little to hold, so it invents.
The fix is not a longer prompt; it is a closer image and a smaller move. The frame rules are in the guide.
- Use a closer reference: head and shoulders rather than a full-body shot. A face a few pixels wide cannot be held.
- Front-facing, evenly lit, eyes visible. Heavy shadow on half the face is where warping starts.
- One face per shot. Two people's references average into one wrong face.
- Keep the motion small. A slow turn costs less face than a fast one.
- Add “no morphing, no extra limbs” to the motion prompt.
Flicker and texture boiling
Fine detail is the hardest thing to keep stable. Hair strands, lace, leaves, fences, stripes and screen patterns shimmer from frame to frame even in a good take: the model redraws the detail slightly each frame, and your eye reads that as flicker.
- Simplify the background: a soft gradient, a blurred street, a plain wall.
- Slow the camera down. A held frame is the most stable shot you can make.
- Add “steady, consistent lighting, no flicker” to the motion prompt.
- Keep fine patterns out of the reference, or crop them before uploading.
- Avoid heavy ambient motion — pouring rain, blowing leaves — when the face matters.
Morphing, drifting and the frame sliding
Morphing is structure changing rather than moving: the background warps around the subject, the outfit becomes another outfit, the scene slides sideways. It usually means the instruction asked for more travel than five seconds can carry, or the still itself is ambiguous — a blurry edge gives the model permission to move it.
The tool has one move per shot and no cuts. If a shot must travel across a location, that is two shots and an edit outside FicReel, not one longer prompt.
- One camera move per shot. Push in or orbit, not both.
- Remove words like snap, fast, whip, quick. Say “slow and steady”.
- Start from a crisp frame; motion blur baked into the still gets amplified.
- If the frame really must travel far, try the longer 10-second duration instead of a faster camera.
- Be explicit: “no cuts, no transitions, no morphing”.
A CALM MOTION LINE TO REPLACE AN AGGRESSIVE ONE
The camera pushes in slowly and steadily, one continuous take. The coat moves in the wind; the background stays still. No cuts, no morphing.
Garbled text, watermarks and logo shapes
Image models draw letter-like shapes, not words. Signs, book covers, printed shirts and UI in the reference come back as plausible gibberish, and the video makes it shimmer. No prompt reliably spells a sentence correctly inside an anime shot.
The practical fix is to remove text from the input. If a word must appear in the frame, add it afterwards in an image editor or video editor, outside the tool.
- Add “no text, no watermark, no logos” to every image prompt.
- Crop printed text out of the reference before uploading it.
- Avoid framing on signs, books, screens and posters.
- Do not try to spell words in the prompt; ask for the object, not the letters.
Extra characters and limbs
A second figure appears at the edge of the frame, or an arm grows out of a coat. Crowded references invite this, because the model reads a background person as part of the subject. It is also common in wide shots, where the body is small and the model has more room to invent.
One subject per reference, cropped tight, with a calm camera, prevents most of it; add “solo character, no other people” when the frame is busy.
A triage order that saves credits
Change one thing at a time, in this order. Each step costs either a frame or a shot, so the cheap checks come first. The two stages are Create for the frame and Create video for the movement.
Judge the frame before the shot
If the still is wrong, no motion prompt will save it. A frame costs one image credit to redo.
Check the reference
Closer, brighter, one subject, clean background. This is the highest-leverage change in the list.
Shrink the motion
Halve the camera move and the subject action before you change anything else.
Rethink the duration
5 seconds cannot carry a long travel; the 10-second option costs 90 credits and is a different take, not a repaired one.
Generate again last
Re-pressing with identical inputs is a lottery; only do it once the setup is right.
| Symptom | Most likely cause | Change first |
|---|---|---|
| The face becomes another person | Reference too small, dark or turned away | A closer, front-lit face reference |
| Hair and patterns shimmer | Fine detail plus camera motion | Simpler background, slower move |
| Background warps or slides | Camera move too aggressive for the length | One calm move, or the longer duration |
| Letters appear on shirts or signs | Text in the reference or the prompt | Add “no text”, crop the reference |
| An extra arm or a second person | Crowded reference, wide framing | One subject, tighter crop, “solo character” |
When to stop retrying
Two or three takes is a working budget for one setup. If the third fails the same way, the problem is upstream — the reference, the length, or the frame — and retries will not fix it. Change the input, not the dice.
A take that is mostly right with one weak moment is normal. If the face and framing are right, the shot has done the hard part; 45 credits for a usable shot beat another 45 chasing a perfect one.
Failed generations are returned automatically. A result you dislike still costs its credits, which is why the order matters: a frame you can reject for 6 credits is cheaper than a shot you reject for 45.
Questions
Common questions
- Why does the face warp only sometimes?
- Every generation is a new take from the same inputs. Reference and move size change the odds, not the outcome — tighten the reference and calm the camera, but no setting makes it 100%.
- Can I fix flicker with a prompt alone?
- Partly. “Steady, consistent lighting, no flicker” helps, but simplifying the background and slowing the camera does most of the work.
- Why does text turn into gibberish?
- These models draw letter-like shapes rather than spell. Keep text out of the reference, add “no text, no watermark, no logos”, and composite real words afterwards.
- Should I switch to the 10-second shot to avoid morphing?
- Not as a general fix. A longer shot gives a slow move room to finish, but it is a separate 90-credit take and can look slightly different. First try one calm move at 5 seconds.
- Do I get credits back when a take is bad?
- Failed tasks are returned automatically. A result you dislike is a completed generation and costs its credits, so judge the frame before you pay for motion.
Try the workflow on one frame.
New devices start with free credits; nothing spends without your press.
FicReel is an AI anime image and video studio. Prices and wait times on this page come from the same credit table the product uses; if the product disagrees with this page, that is a bug — write to admin@mail.ficreel.com. More: beginner guide, pricing, all guides, glossary, about FicReel. Canonical source: https://ficreel.com/learn/ai-anime-video-failures-and-fixes.