First, clarify that the acceptable flaw tolerance stems from communication goals, not technical capabilities.
When evaluating AIGC commercials, the most common mistake brands make is treating the technical ceiling as the quality standard. Unstable details in AI-generated imagery, occasional limb deformations, and text rendering errors are phenomena that can almost never be completely eliminated. Acceptable flaw tolerances should be reverse-engineered from communication goals, rather than forward-engineered from generation results.
The initiation phase needs to answer three questions. First, on which platform will this video be placed, an outdoor large screen or a mobile feed, as screen size and viewing distance determine how many details the audience can perceive. Second, what is the core message; if the selling point is the product's appearance, then the accuracy of the product's form is a hard metric that cannot accept any deformation. Third, does the emotional atmosphere take priority over realism; if the video pursues a surrealist style, then slight deviations from the laws of physics can actually become part of the style.
Specifically, the brand must clearly state non-negotiable and negotiable items in the creative brief. For example, the product logo must not be distorted, colors must be accurate, and character faces must not have obvious deformities, but the number of leaves on background plants can be random. Only after receiving this list can the production team determine which shots need to be redone and which can be kept.
The standard is whether the flaw affects the delivery of core information. A blurry background will not cause the audience to misunderstand the product's function, but a distorted product outline will directly damage brand trust. The risk is that if the brand does not define this in advance, final acceptance will rely solely on subjective feelings, leading to repeated revisions and a loss of control over the project timeline.
An exception is if the project itself is an experimental art short film where the communication goal is to explore visual boundaries; in this case, the tolerance for flaws can be relaxed, and AI generation artifacts can even be deliberately retained. However, this situation requires the brand to reach an internal consensus, otherwise the marketing and creative teams will each stick to their own views.
Write flaw classifications into the storyboard rather than leaving them to the shoot.
A storyboard is not just a visual description, but an operational manual for flaw tolerance. The production team must mark the key areas and acceptable margin of error for each shot during the storyboard phase. For example, in a close-up shot, the character's hand movements must be natural, while slight distortion is allowed in the blurred background areas.
The brand needs to provide reference images, including competitor samples, historical brand assets, and mood boards. These reference images are primarily used to define the visual baseline. Based on the reference images, the production team marks in the storyboard which elements are brand assets that must be accurately reproduced, and which are decorative content that can be freely generated.
Specifically, a "flaw tolerance" column must be added to every shot in the storyboard, divided into high, medium, and low levels. High means obvious AI artifacts are acceptable, medium means only slight detail deviations are accepted, and low means precision close to live-action or CGI is required. The brand and the production team jointly confirm the level for each shot, sign, and file it.
The standard is that the closer a shot is to the product's core selling point, the lower the flaw tolerance. For example, in a product promotional video, the tolerance for product close-ups must be set to low, while environmental B-roll can be set to high. The risk is that if this is not clarified during the storyboard phase, on-set adjustments will waste a significant amount of generation time.
An exception is if the brand wants to use AI to generate unpredictable, accidental effects; in this case, certain shots can be deliberately set to high, but it must be agreed in advance which shots can be experimental and which must be conservative. Otherwise, if an experimental shot fails, there will be no backup plan.
Use traditional film and television workflows to constrain AIGC assets during the production phase.
AIGC commercials are not entirely detached from filming. Many projects require shooting live-action footage first, followed by AI generation or compositing. The core task during the shooting phase is to collect clean, stable base footage, leaving redundancy for post-production AI processing.
Brands need to provide resources such as locations, actors, and props, and confirm shooting conditions. The production team must test the AI generation model's compatibility with live-action footage in advance, such as whether the resolution, frame rate, and color space of the footage meet the model's requirements. If the model requires a specific format, camera settings must be adjusted during shooting.
Specifically, the shooting set must record additional multi-angle reference images, including multiple angles of the product, actors, and scenes. These reference images are used to calibrate details during post-production AI generation. At the same time, shooting should avoid factors that easily trigger AI errors, such as reflections, occlusions, and motion blur.
The standard is that the higher the quality of the live-action footage, the lighter the burden of post-production AI restoration. If the live-action footage itself has noise or is out of focus, the AI generation results will amplify these defects. The risk is that brands may compress shooting time to save costs, resulting in insufficient footage and the inability to generate ideal visuals in post-production.
An exception is that if the project relies entirely on AI generation and requires no live-action shooting, the shooting phase can be omitted, but more time is needed for model training and prompt debugging. At this time, the tolerance for flaws should be set more loosely because the controllability of the generated results is lower.
Establish a frame-by-frame inspection and flaw repair process in the post-production phase.
Post-production is the phase where flaws in AIGC commercials are most concentrated. Brands and production teams need to jointly establish a frame-by-frame inspection process rather than just looking at the overall effect. AI-generated videos often exhibit obvious errors at certain moments, such as sudden changes in the number of fingers, flickering text, or jittering object edges.
Specifically, the post-production team first outputs a low-resolution preview, and the brand marks all visible flaws on the preview, grading them by severity. Severe flaws include product deformation, abnormal facial features, and text errors, which must be fixed. Minor flaws include unnatural background textures and slight lighting inconsistencies, which can be accepted or masked by color grading.
The standard is that if a shot contains more than three severe flaws, it is recommended to regenerate it. If the cost of regeneration is too high, local repair tools can be used, but the repair traces must not affect the overall style. The risk is that the brand may demand repairs for every flaw, leading to infinitely rising costs, making early grading crucial.
Post-production must also focus on audio and subtitles. AI-generated voiceovers may sound mechanical and require human dubbing or adjustment. The font, position, and timing of subtitles must be synchronized with the visuals, as any misalignment will appear unprofessional. Brands must provide subtitle copy and style requirements, and participate in subtitle proofreading.
An exception is that if the project is for internal testing or non-public use, the tolerance for flaws can be relaxed, provided the core information remains clear. However, for official releases, it is recommended to reserve time for at least two rounds of revisions, outputting a new preview for confirmation after each round.
The acceptance checklist must be broken down by stage, rather than just reviewing the final video.
Acceptance for commercial video projects cannot be limited to delivering a single final video file. The brand must conduct acceptance checks separately for the script, storyboard, shooting materials, editing, color grading, audio, subtitles, master, and source files. Each stage has its own independent standards for flaws.
Specifically, the production team should provide an acceptance checklist upon delivery, listing all deliverables and their formats. The brand checks each item against the list. For example, whether the script contains the final copy, whether the storyboard matches the final video, whether the shooting materials are fully archived, and whether the color-graded version complies with brand color specifications.
The standard is that the acceptance criteria for each stage must be written into the contract at project launch. The brand cannot wait until the final video stage to request changes, as the cost of reworking later is extremely high. The risk is that many projects only accept the final video and ignore the source files, making subsequent revisions impossible.
During acceptance, the scope of copyright must also be considered. Copyright ownership of AI-generated materials may involve model training data, so the brand needs to confirm the scope of use, including whether it can be used for advertising, modified, or transferred. The production team must provide generation logs and model versions for legal verification.
An exception is that if the project is a rapidly iterating social media short video, acceptance can be simplified to checking only the final video and subtitle files. Even so, source files must be retained because platform rules may change, requiring different versions to be re-edited.
When the scope of flaws is inapplicable, cut losses promptly or adjust the plan.
Some projects are unsuitable for AIGC production, or current technology cannot meet the requirements for flaws. The brand must identify these situations during the project initiation stage to avoid wasting budget.
Inapplicable conditions include products with extremely complex appearances that require precise display, such as car interiors and precision instruments. AI generation is prone to detail errors, and the cost of fixing them is high. Additionally, if a brand has strict visual specifications requiring every pixel to meet standards, AIGC may not be able to satisfy them.
Specifically, before project kickoff, the production team can run a quick test by generating a few key shots with AI, allowing the brand to evaluate if the flaws are acceptable. If the test results fall short, it is recommended to switch to traditional CGI or live-action shooting. The testing cost is far lower than the cost of reworking later.
The criterion is that if severe flaws appear in more than half of the test shots and cannot be reduced by optimizing prompts, the project risk is too high. The risk lies in the brand potentially forcing the project forward due to time pressure, ultimately delivering poor results that damage the brand image.
An exception is if the project itself is a proof of concept or internal demo not meant for public release; in this case, the tolerance for flaws can be significantly relaxed, and even obvious AI artifacts can be accepted to showcase the creative direction.
The next recommendation is to conduct a small-scale test before deciding on the full video.
For brands currently evaluating AIGC commercials, it is recommended not to launch full production directly. First, select a key shot or a 15-second clip to produce a test version, clarify the tolerance for flaws, and establish acceptance criteria. The test version can be used for internal discussions or presented to executives and clients.
During the testing phase, generation parameters, the repair process, time spent, and costs must be recorded, as this data helps determine the feasibility and budget for full production. If the test results meet expectations, expand to the complete project. If the test is unsatisfactory, the creative direction can be adjusted or other production methods adopted.
At the same time, it is recommended that brands sign phased contracts with the production team, with clear deliverables and acceptance criteria for each phase, to avoid excessive one-time investment. This way, even if issues are discovered midway, losses can be stopped in time.
Finally, stay updated on the development of AI technology. Generative models update rapidly, and flaws that are unacceptable today might be resolved next month. However, project decisions cannot rely on the future and must be based on current capabilities.
If you are preparing an AIGC commercial project, you can first organize your brief, reference visuals, product or company materials, delivery platforms, and copyright scope, and then view theAIGC video services page., grounding communication from abstract preferences into actionable production boundaries.
