Confirm the role real product frames can play before project initiation.
A common misconception in AI product videos is treating AIGC as a shortcut to replace shooting. Real product frames serve as visual anchors in a project, not a pile of materials. Brands must first provide three to five live-action frames that best represent the product's form, material, lighting, and interaction logic. These frames will determine the style baseline for all subsequent generative visuals.
The standard is that each reference frame must answer three questions. First, are the product's scale and angle in the frame clear? Second, do the ambient light and surface reflections conform to real physical logic? Third, are there brand logos or key functional details in the frame, and do these details need to remain recognizable during AIGC generation? If the reference frames are blurry or have limited angles, structural drift will occur in the subsequent generated visuals, leading to product distortion or loss of features.
The action required during the project initiation phase is to establish a reference frame archive. The archive should include original frames, cropped composition frames, color tendency annotations, material descriptions, and restricted modification areas. The production team tests the generative model based on the archive and outputs three style test images. The brand determines whether the model understands the product's core features based on the test images, rather than just evaluating if the visuals look good.
The risk is that reference frames may contain copyrighted patterns or undisclosed design details. The brand should clear non-essential information from the frames in advance, such as third-party trademarks in the background, unreleased accessories, or people's faces. If not cleared, the generative results may inadvertently copy these elements, leading to legal issues upon delivery.
An exception is when the product itself is in the concept phase and no real frames are available. In this case, product design drawings, 3D wireframes, or physical prototype photos should be used as alternative references, but the hierarchical differences of the reference sources must be noted in the project initiation documents to avoid disputes over the product's form during later acceptance.
How to design shots around real product frames during the shooting phase.
The focus of shooting is to provide usable base footage for AIGC post-production. The production team needs to plan the camera movement trajectories, shot sizes, and focus shifts for the live-action shots based on the product angles in the reference frames. Before each shot is filmed, a camera position diagram should be output and compared side-by-side with the reference frame to confirm that the composition ratio and visual center of gravity are consistent.
Specific actions include using a monitor on set to overlay a semi-transparent layer of the reference frame, allowing the camera operator to adjust the gimbal angle and lens focal length accordingly. The gaffer needs to reference the highlight positions in the reference frame to set up the key light, fill light, and rim light, making the product's surface reflections match the reference frame as closely as possible. If the reference frame features outdoor natural light but the shoot is completed in a studio, large softboxes should be used to simulate skylight, and environmental reflection details should be supplemented later via AIGC.
The standard is that each shot must retain at least three seconds of stable footage, and the product must occupy no less than 80% of the frame compared to the reference frame. Below this ratio, it is difficult to preserve product texture details during post-production generation. Additionally, the shot footage should include a set of empty shots without products or people, used for generating environmental extensions or replacing backgrounds in post-production.
The risk is an excessive color temperature difference between the live-action footage and the reference frame. For example, if the reference frame is cool and the live-action footage is warm, AIGC may force color correction during generation, resulting in product color distortion. The solution is to calibrate the monitor with a color chart before shooting and to record one minute of gray card footage before each scene for use in post-production color matching.
An exception is when the product is too large or too small to be shot from the reference frame angles. Large equipment can be shot in segments and stitched together in post-production, while small products should use macro lenses paired with motorized turntables. In this case, the role of the reference frame is to constrain the proportional relationships of the final composite image, rather than to serve as a frame-by-frame unified standard.
Using real product frames to constrain generative visuals in the post-production phase.
Post-production is the phase where AIGC is most deeply integrated, but real product frames must serve as constraints throughout. The production team should input reference frames into the generative model as condition maps while setting product feature preservation weights. Excessively high weights will result in stiff visuals, while excessively low weights will cause product distortion. It is recommended to output comparison sequences with different weights during the testing phase, allowing the brand to select an acceptable balance.
The specific execution process involves first generating static style frames from reference frames, and then generating dynamic shots after confirming the product's appearance and materials are correct. Dynamic shots require frame-by-frame inspection to check for flickering or distortion along the product edges. If flickering occurs, the generation frame rate should be reduced, and frame interpolation algorithms should be used to restore smoothness. If distortion occurs, return to the reference frame to re-annotate the product outline and add an outline constraint layer.
During the editing phase, AIGC-generated shots must be mixed with live-action shots, but differences exist in sharpness, noise, and motion blur. The production team should use matching plugins to unify the texture and insert product close-up frames at transitions as visual buffers. When reviewing the video, the brand should focus on whether the product feature demonstration is clear, such as whether button positions, port orientations, or screen content match the real product.
The color grading phase uses the reference frame's color palette as a baseline, applying the same LUT to both generated and live-action shots before manually fine-tuning highlights and shadows. Sound design must synchronize with product actions in the visuals, such as rotation, opening and closing, or touch feedback; these audio details enhance realism and compensate for potential physical unnaturalness in AIGC visuals.
The risk is over-relying on AIGC in post-production, causing product features to be obscured by stylized filters. The brand should require the production team to retain at least one version of the original generated sequence without filters for comparative acceptance. If product logos are altered or structures are simplified, it should be deemed unqualified and require regeneration.
An exception is when the product itself has highly reflective surfaces, such as mirrors or glass. AIGC generation is prone to environment mapping errors in these cases; therefore, live-action product frames should be used as the foreground layer, with AIGC handling only the background and lighting effects to avoid directly generating the main product.
The acceptance checklist should be divided by phase to avoid evaluating only the final video.
Acceptance cannot rely solely on the final video; it requires separate documentation for the script, storyboard, shooting materials, generated sequences, edited versions, color-graded versions, sound mixes, subtitles, masters, and source files. Each phase should have independent passing criteria, jointly signed and confirmed by the brand and the production team.
Script acceptance must confirm whether product selling points are accurately conveyed and whether key actions agreed upon in the reference frames are included. Storyboard acceptance requires comparing the composition of each frame with the reference frames, with deviations exceeding 15% deemed unqualified. Shooting material acceptance must check for sufficient product close-ups and B-roll; if insufficient, reshoots should be arranged rather than forcefully filling gaps with AIGC in post-production.
Generated sequence acceptance requires shot-by-shot inspection of product feature preservation, edge stability, color consistency, and dynamic smoothness. Edited version acceptance must confirm narrative pacing, transition logic, and brand information exposure duration. Color-graded version acceptance requires comparing the reference frame color palette on a standard monitor, with allowable deviations remaining within a visually acceptable range.
Audio mixing acceptance must check whether product action sound effects are clear and whether background music overpowers the voiceover. Subtitle acceptance must verify that the text content matches the product name and functional terminology, and confirm there are no typos. Master and source file acceptance must confirm that the delivery format meets the publishing platform's requirements, with platform specifications based on the latest official requirements prior to release, while retaining editable project files.
The risk is that the brand only accepts the final video and ignores intermediate steps, causing post-production modification costs to multiply. For example, if a composition deviation is not discovered during the storyboard phase and is only exposed during the generation phase, model parameters must be readjusted and regenerated, causing the schedule and budget to spiral out of control. Therefore, each step should have clear time milestones and a sign-off process.
An exception is when the project cycle is extremely short, making step-by-step acceptance impossible. In this case, priority should be given to accepting the product feature frames and the final video, while the production team conducts internal quality checks for the remaining steps and issues a written report, allowing the brand to raise objections within seven days after delivery.
Situations where AIGC commercials are unsuitable must be identified in advance.
Not all brand projects are suitable for AIGC commercials. When a product needs to display precise internal structures, authentic material textures, or strict physical interactions, AIGC-generated visuals may not achieve industrial-grade accuracy. For example, medical devices, car engines, or high-end watches are categories better suited for traditional live-action shooting or CGI modeling.
When a brand requires a large amount of authentic user feedback or usage scenarios, AIGC-generated characters and scenes may lack credibility. Audiences can easily identify the stiff expressions of virtual characters or abnormal lighting in scenes, which instead weakens brand trust. In this case, live-action footage should be used, or AIGC should only be used to assist with background extension.
When the project budget is extremely low and the timeline is tight, the testing and iteration costs of AIGC may exceed those of traditional shooting. Because generative models require multiple parameter adjustments and re-renders, each test consumes time and computing power. Brands should evaluate whether they have a sufficient internal time window; otherwise, they should choose a more reliable shooting solution.
When a brand requires strict compliance reviews, such as for pharmaceutical, financial, or children's products, AIGC-generated content may struggle to pass regulatory audits. Auditors typically require proof of original shooting or product authenticity statements, which AIGC visuals cannot satisfy. In this case, generative imagery should be avoided.
The criterion is whether the project meets one of the following conditions: the product has clear visual features that can be constrained by real frames, the brand is willing to invest time in multiple rounds of testing, and the final delivery platform does not require proof of authenticity. If these are not met, it is recommended to adopt a traditional production workflow, or only use AI tools in the post-production color grading and visual effects stages.
Checklist of materials the brand needs to prepare.
The brand should prepare six types of materials before the project starts. First are high-resolution, multi-angle product photos, including front, side, back, and detail close-ups. Second are product design source files or 3D models, used to supplement missing angles in reference frames. Third are brand visual guidelines, including the logo, standard colors, fonts, and prohibited elements.
Fourth are target audience descriptions and distribution scenario explanations, such as placement platforms, viewing devices, and expected viewing duration. Fifth are competitor videos or industry benchmark cases to help the production team understand the style direction. Sixth is a product feature list detailing the core selling points and operational actions to be showcased in the video.
Each type of material should be labeled with a version number and update date to prevent the production team from using outdated assets. The brand should designate a point of contact responsible for answering detailed product questions and providing written feedback after each review. Feedback should be specific to timecodes and visual descriptions, such as noting the product rotation angle is incorrect at the 15-second mark, rather than saying the overall feel is off.
The risk is that incomplete materials cause frequent rework by the production team. For example, missing photos of the product's bottom may result in a blurred bottom structure in the generated footage, which the brand only discovers during final acceptance. Preparing materials in advance reduces ineffective communication, but the production team should also proactively identify material gaps early on rather than waiting passively.
An exception is when the product is still in development and its appearance may be adjusted. The brand should provide actual frames of the current version and clearly indicate which areas are subject to change. The production team should specify in the contract that significant changes to the product's appearance after shooting will incur additional modification fees.
Key actions the production team needs to confirm.
In the pre-production phase, the team should complete technical tests and output a generated style sample along with a product feature retention test report. The test report should include result comparisons under different model parameters and the choices the brand needs to make. The team should also confirm the copyright ownership of reference frames; if they come from third-party asset libraries, authorization proof must be obtained.
A technical coordination meeting should be held before shooting to clarify the division of labor between live-action and AIGC. This includes determining which shots are entirely live-action, which are generated by AIGC, and which are composites of live-action and generated elements. The team should output a shot breakdown chart detailing the production method and acceptance criteria for each shot. This chart serves as the basis for post-production acceptance and is key to cost control.
During post-production, the team should implement a version control system, retaining the version number and modification notes for every revision. When the brand reviews the video, the team should provide comparison versions, such as the differences between the previous and current versions, to help the brand make quick decisions. The team should also sync progress regularly to avoid a drop in quality caused by concentrated feedback at the final stage.
Upon delivery, the team should provide a complete source file package, including generated sequences, compositing projects, color-graded versions, audio stems, and subtitle files. The source file package should include a documentation file recording the purpose and modification method for each file. The brand should verify that the source files can be further edited by other teams, rather than being exclusively dependent on the original production team.
The risk is that the team skips technical testing and goes straight into formal production, only to discover later that the model cannot handle the product features, requiring retesting. This causes a double loss of time and budget. The team should clearly define the workload and deliverables for the testing phase in the contract, ensuring the brand pays for the testing rather than treating it as a free service.
An exception is when the brand already has a long-term production team that is familiar with the product line. In this case, the testing process can be simplified, but the reference frame files and shot breakdowns must not be omitted, or communication discrepancies will still arise later.
The next step is to start with a minimum viable test.
If a brand is interested in AIGC commercials, it is recommended to first select one product and produce a fifteen-second test video. The test video does not need a complete narrative; it only verifies product feature retention and style matching. Keep the testing cycle within two weeks and the budget under ten percent of the total project budget. After testing, decide whether to proceed with the formal project based on the results.
The test video should include three shots: a product close-up, a usage scenario, and a dynamic transition. Each shot must use real product frames as references and output a live-action comparison. The brand and the production team jointly evaluate the test results, recording what meets the standards and what needs improvement. Evaluation criteria include product recognizability, visual fluidity, brand style consistency, and delivery format compatibility.
If the test results meet the standards, proceed to formal production and confirm each step according to the aforementioned acceptance checklist. If the test results are unsatisfactory, analyze whether the issue lies with the reference frames, model parameters, or shooting materials, make targeted adjustments, and test again. Do not give up on AIGC because of a single failure, nor blindly expand its use because of a single success.
Finally, it is recommended that brands clearly stipulate the copyright ownership and revision limits for AIGC-generated content in the contract. Copyright rules for generative video are still being refined, and early agreements can prevent future disputes. At the same time, brands are reminded to pay attention to the labeling requirements for AIGC content on publishing platforms, subject to the latest official platform regulations.
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 pageto ground communication from abstract preferences into executable production boundaries.
