Section 3: Image Generation and Visual Language · Lesson 4 of 5 (14 of 45) · 60 minutes · Evidence: Artifact diagnosis sheet
A generator never says “sorry, that hand has seven fingers.” It hands you a polished image and lets you believe it is finished. Audiences notice flaws they cannot name: a sign that spells nothing, a ladder that does not touch the ground, a face that is slightly two different faces. Today you become the quality-control department for your own film.
Production journal: write your answer below.
What must a filmmaker understand before they can identify anatomy text object perspective and background errors and choose a repair method?
An artifact is any visible flaw introduced by the generation process: something wrong, odd or unintended that you did not ask for. Artifacts are normal. Finding them is part of the job, and the filmmaker is the only one who can, because the tool cannot see its own mistakes. This habit is called quality control: checking every frame against the same list before it is allowed into the film.
Look at each frame at full size and then slowly, one region at a time.
Choose the cheapest method that works, in this order.
Perspective problems, such as a floor that tilts the wrong way, are usually fixed by regenerating with a clearer camera description, because they run through the whole picture.
In a frame for a detective comedy, the student finds three problems. A shop sign shows near-words (text error). The detective’s left hand has six fingers (anatomy). A coat rack on the right seems to grow out of the wall (object). The student crops the shot slightly to remove the coat rack, inpaints the hand, and writes “no readable text, blank sign” into the prompt to regenerate the sign area. After two tries the hand is still wrong, so the student switches to a different take in which the hand is in a pocket. Every step goes on the diagnosis sheet.
Image editors and compositors at professional studios treat every incoming image as unfinished until it has passed review. Creators who rely on generators follow a similar routine, usually informally. Study the routine and decide at each step whether judgment or machinery is at work.
The reviewer begins by viewing the whole frame, then zooms into faces, hands, the edges of objects and the background. They look for any text, because writing is among the most common failures. When they find a flaw, they decide whether the frame is worth repairing, based on how close the flaw is to the subject and whether the audience would notice it in motion. A small background flaw on a shot that will last one second may be acceptable; a flaw on a face in a close-up is not.
They then try the lightest fix first, and record what they did and whether it worked. The tool does the redrawing, but only the reviewer decides when to stop repairing and start over.
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 |
|---|---|---|---|
| Inspect the frame | Views full frame, then zooms into hands, faces, text, objects and background | Nothing; it cannot judge its own mistakes | That no flaw was missed, including unintended readable text and symbols |
| Diagnose the artifact | Names the flaw and judges how much it would distract the audience | Nothing | Whether the flaw is acceptable in a short or distant shot |
| Choose a repair method | Selects crop, inpaint, regenerate or reject, lightest first | Redraws the selected area or the whole image | That the repair did not create new flaws or change the character |
| Record the decision | Writes the method, the result and the reason | May log settings | That the record is accurate and complete |
Studio task: Annotate flawed images and match each problem to regenerate edit crop or reject.
Your teacher will supply three to four flawed sample images (or you may use flawed outputs you generated yourself). Work in pairs.
Coaching tips. Do not accept “it looks off” as a diagnosis; name the exact part and the category. If the flaw is on a face or touches a story point, lean toward reject. If your tool does not offer inpainting, pretend the option is unavailable and decide what you would do instead.
Frame ID: ______ Source / tool / date: ______ Flaw location (describe the part of the frame): ______ Category: anatomy / text / object / perspective / background / style drift / unsafe content How distracting (1 low - 3 high) and why: ______ Repair method: crop / inpaint / regenerate / reject What I changed (prompt words, area, settings): ______ Result: fixed / partly fixed / failed -> next step: ______
Studio task: Repair or replace two failed project frames and document the method.
Work on failed frames from your own project, or flawed frames from your prompt tests in lesson 3.2.
Checklist before you submit:
Name the artifact that would most distract the audience.
Deliverable: Artifact diagnosis sheet.
Submit the artifact diagnosis sheet. 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_S03-L04_artifact-diagnosis-sheet_v01_YOURINITIALS
| Criterion | Advanced | Proficient | Developing | Beginning |
|---|---|---|---|---|
| The work addresses the stated story or production purpose. | Repaired or replaced frames serve the intended shot purpose and the sheet shows why each choice protected the story. | Repaired frames match the shot plan and story purpose. | Frames only partly match the shot plan. | Frames do not connect to the project. |
| The student applies the required method accurately and produces usable capstone evidence. | Flaws are named precisely by category, methods are chosen lightest-first, and rechecks are documented. | Flaws and methods are accurate and mostly complete. | Some flaws are missed or methods do not match the problem. | Diagnosis is vague or missing. |
| The student identifies at least one quality issue and records a meaningful revision decision. | Records what changed between attempts and a clear decision to repair or reject. | Identifies at least one quality issue and records a meaningful revision decision. | Notes a problem but gives no clear revision. | No revision decision recorded. |
| The work complies with consent rights attribution privacy disclosure and school policy. | Screens every frame for unsafe symbols, unintended identities and text, documents each rejection and cites sources. | Screens for unsafe content and records sources. | Screens partially or documents few sources. | Keeps unsafe content or unintended likenesses. |
Requirement: Reject outputs that introduce harmful symbols unsafe content or unintended identities.
Quality control includes a safety check. Generators sometimes add symbols, slogans, body shapes or faces that resemble real people or carry harmful meaning. Reject any frame with a harmful symbol, unsafe content or an identity you did not intend, and record the rejection so your teacher can see the check was done. Never repair a frame by pasting in a real person’s face or a protected logo. Keep a hand-drawn or photographed alternative ready in case a tool will not produce a safe result.