Section 4: Character Consistency · Lesson 5 of 5 (20 of 45) · 60 minutes · Evidence: Approved six-frame consistency grid
A studio does not release a character into production because someone feels good about it. It releases it because the character passed a test that someone else could repeat. Today you build that test: six frames, a checklist with clear pass or fail lines, and a written decision to approve the character, or send it back.
Production journal: write your answer below.
What must a filmmaker understand before they can evaluate six frames against measurable identity and continuity criteria?
“It looks about right” is not a quality standard, because two people will disagree. A consistency grid is a six-frame test that puts your character in a deliberate range of conditions and records the results in one place. It answers a single question: can this character be trusted in the shots my film needs?
Choose six frames that stress different things, so one weakness cannot hide. A good set is: (1) front, neutral; (2) three-quarter turn; (3) profile or back; (4) a strong expression; (5) a different setting or lighting; (6) a wide shot where the face is small. If your film has two main characters, make at least one frame a pair. Use your fixed wording and unchanged identity block in every prompt.
Acceptance criteria are the specific conditions a frame must meet to count as a pass. Write them before you generate, so you cannot bend them to fit what you get. Make each criterion a yes or no question. For example:
A frame passes when every applicable question is yes. Set a bar for the whole grid too, such as “five of six frames pass, and no failed frame is missing an anchor.”
A revision is a deliberate change made because of a named failure, followed by a retest. Change one thing, regenerate the failed frame, rescore. When the grid meets your bar, write an approval note: the date, which frames passed, which required revision, what the known weak spot is, and who signed off (you and a second reviewer, such as a peer or teacher). Approval does not mean perfect. It means the remaining risk is known and acceptable.
A student tests Kofi, a sailmaker with a braided beard and orange vest. Frames 1 to 4 pass. Frame 5, in a dim workshop, loses the orange vest to a brown one: fail on wardrobe. She adds “bright orange vest” to the lighting line, regenerates, and it passes. Frame 6, wide, has a blurred face, which she notes as a risk and decides to avoid wide shots of Kofi unless he is near the camera. Five of six pass; she approves with that note.
Quality checks in production work only when they are separate from the person who made the image. Study how an experienced creator tests a character and watch which parts need a second pair of eyes.
The creator writes the pass or fail questions first, then generates the full set in one sitting with unchanged wording, so that differences come from the character rather than from the prompt. Each frame is scored against the canonical reference, not against memory. Failures are classified by cause, such as a lost anchor, a lighting problem or an artifact, and the creator fixes only the failed frames. The rejected versions are kept in the log because they show what went wrong and justify the changes. A second reviewer then scores independently; where the two scores disagree, the criterion was probably unclear and gets reworded.
The tool produces the frames. Only people can judge whether a result is acceptable for this film.
As you study, record each decision point in your production journal.
| Decision point | What the filmmaker does | What the tool automates | What still needs human review |
|---|---|---|---|
| Write acceptance criteria | Turns quality into yes-or-no questions and sets the grid’s pass bar | Nothing | That the criteria match what this film really requires |
| Generate the six frames | Runs all frames with unchanged wording in one sitting | Draws each frame | That the frames stress different conditions and no weakness can hide |
| Score and classify | Scores each frame against the canonical reference and names each cause of failure | Nothing | Whether a second reviewer scores the same frames the same way |
| Revise and approve | Changes one thing per failure, retests and writes the approval note with a known risk | Regenerates failed frames | That the fix did not break a passing frame, and the remaining risk is acceptable |
Studio task: Score sample frames with the consistency rubric and calibrate judgments.
Work in groups of three, using the sample six-frame grid your teacher provides, with a canonical reference and a rubric. If you have your own grid in progress, you may score a classmate’s instead.
Coaching tips. Score against the canonical reference image, not against how you remember the character. If your group splits on a frame, ask which specific anchor or detail is in dispute and look at it side by side. Resist rounding up: a half-pass is a fail until a criterion says otherwise. Practise writing the approval note in plain words that a producer who has never seen your project could act on.
Acceptance criteria (yes/no): anchors present | face and hair match | wardrobe and colours | age and build | no artifacts | (pairs) scale, sides, eyelines Grid pass bar (e.g. 5 of 6, no missing anchor): ______ Frame | What it tests | Criteria passed | Pass/Fail | Cause if failed 1 | front neutral | ______ | ______ | ______ 2 | three-quarter | ______ | ______ | ______ 3 | profile or back | ______ | ______ | ______ 4 | strong expression | ______ | ______ | ______ 5 | new setting or lighting | ______ | ______ | ______ 6 | wide shot / pair | ______ | ______ | ______ Revisions made (one change each, with retest result): ______ Approval note: date ______ | frames passed ______ | known weakest risk ______ | signed off by ______
Studio task: Complete the six-frame test revise failures and request production approval.
Run the full six-frame test on your main character (or on your two-character pair), and request approval for production.
Checklist before you submit:
Record the weakest remaining continuity risk.
Deliverable: Approved six-frame consistency grid.
Submit the approved six-frame consistency grid. The file must use the course naming convention and include enough process evidence to show the student's decisions and revisions.
File name (course naming convention): AIF_S04-L05_approved-six-frame-consistency-grid_v01_YOURINITIALS
| Criterion | Advanced | Proficient | Developing | Beginning |
|---|---|---|---|---|
| The work addresses the stated story or production purpose. | The grid tests the exact shots the film needs and the approval note states what production it clears the character for. | The grid tests the shots the film needs and the approval is clear. | The grid is generic with limited link to the film. | The grid does not relate to the film. |
| The student applies the required method accurately and produces usable capstone evidence. | Criteria were written first as yes-or-no questions; six varied frames are scored accurately with causes and a second reviewer. | Criteria are clear, frames are varied and scoring is accurate. | Criteria are vague or frames are not varied enough. | No criteria, or the frames are not scored. |
| The student identifies at least one quality issue and records a meaningful revision decision. | Revises each failure with one targeted change, retests, and names a remaining risk with a plan for it. | Records a revision for at least one failure and names a risk. | Notes a failure but no revision. | No revision or risk recorded. |
| The work complies with consent rights attribution privacy disclosure and school policy. | Process log keeps rejected outputs, discloses AI use and documents provenance and permissions for every image. | Process log and disclosure are present and sources are recorded. | Some outputs are missing or disclosure is incomplete. | Rejected outputs are removed or sources and disclosure are missing. |
Requirement: Keep rejected outputs in the process log when they explain a meaningful correction.
Your log is part of your honesty as a filmmaker. Keep the rejected frames when they explain a correction, record which images were made with an AI tool, and say so on the final grid. Keep the provenance for every reference image and any signed releases. A second reviewer should be a person, not the tool that generated the images. If a frame passes only because a tool cleaned up a flaw you could not fix yourself, say that in the note, and do not hide it. Provide alt text for each frame so the grid is usable by classmates who cannot see images.