MiniMax H3 reference inputs: a 16-case readiness audit

Input roles

Four image slots are linked. The prompt describes them as successive visual states; a generic reference_image label alone would not explain the sequence.

Missing files and open checks

• The source does not declare requiredReferenceCount; the number of listed slots is not a provider limit.

4 listed slots · 4 linked · 0 unlinked

Why it is included

Included to show why a complete URL list still needs a role inventory. Lens-mask and lettering requests are source instructions, not verified results.

Next preparation step

Prepare four of your own images, label their intended order and separate any exact lettering into an editable overlay. Confirm the chosen provider accepts this reference arrangement before submission.

Input roles

Image 1 supplies the overall setting/style, Image 2 the person, Image 3 the bag and Image 4 the ending logo according to the source prompt.

Missing files and open checks

• The source does not declare requiredReferenceCount; the number of listed slots is not a provider limit.

4 listed slots · 4 linked · 0 unlinked

Why it is included

Included because equal file types can carry different jobs. Four reachable JPEGs do not establish rights to a person, product or logo.

Next preparation step

Write an identity/product/style/ending-role sheet for your own assets. Keep a precise brand end card in post-production instead of assuming the model will print a logo correctly.

Input roles

The raw originalMode is text and the references array is empty. That is an explicit text-labeled source record, rather than an unspecified-mode record with no listed files.

Missing files and open checks

• The source does not declare requiredReferenceCount; the number of listed slots is not a provider limit.

0 listed slots · 0 linked · 0 unlinked

Why it is included

Included as a contrast for the unknown-mode cases. The metadata does not prove a current provider setting, input limit or a reproduced laundromat result.

Next preparation step

Use the source page to study how scene and motion are described. For your own attempt, choose a provider-supported text workflow and define one visible action to inspect; do not invent missing source settings.

Input roles

Image 1, Image 2 and Image 3 have links; the source explicitly calls for Video 1 as an editing/rhythm reference, but its reference_video slot has no URL.

Missing files and open checks

• Video 1 has no declared file URL.
• The source does not declare requiredReferenceCount; the number of listed slots is not a provider limit.

4 listed slots · 3 linked · 1 unlinked

Why it is included

Included as a partial bundle. Three accessible images do not replace the absent timing video; copying the prompt would leave a named input unresolved.

Next preparation step

Exclude this bundle from direct replay. Supply an authorized timing clip of your own or rewrite the task as a different image-led exercise and disclose that change.

Input roles

The single linked reference_video is the footage to edit. The prompt asks to replace a desk/background while retaining motion and spatial relationships.

Missing files and open checks

• The source does not declare requiredReferenceCount; the number of listed slots is not a provider limit.

1 listed slots · 1 linked · 0 unlinked

Why it is included

Included to distinguish an editable source clip from a loose style reference. The source wording asks for exact preservation; this audit does not verify that preservation.

Next preparation step

Check whether your chosen provider offers the required video-editing input, then use an authorized clip and define unchanged-object checks. A text/image-only generator is not an equivalent route.

Input roles

The record explicitly has first_frame Image 1 and last_frame Image 2, with a URL for each; originalMode is first-last-frame.

Missing files and open checks

• The source does not declare requiredReferenceCount; the number of listed slots is not a provider limit.

2 listed slots · 2 linked · 0 unlinked

Why it is included

Included as an actual paired-input record, not a rule inferred from a title. It does not prove that the current site/provider accepts a tail-frame slot.

Next preparation step

Inspect both supplied states and confirm a real end-frame control before choosing this route. If it is unavailable, keep the final UI as an editorial end card and describe the exercise as a different workflow.

Input roles

Six reference_image slots and one reference_video slot have URLs. The prompt assigns image content and asks Video 1 to supply shot rhythm, transitions and music.

Missing files and open checks

• The source does not declare requiredReferenceCount; the number of listed slots is not a provider limit.

7 listed slots · 7 linked · 0 unlinked

Why it is included

Included because a larger input bundle has more assignments to check, not because more inputs are better. A reachable clip does not grant music rights.

Next preparation step

Create a seven-slot manifest before trying anything. Verify the provider supports the actual mix and count; replace music with an authorized track and keep timing notes separate from identity notes.

Input roles

Image 1 and Video 1 are both linked. The prompt uses the still for the pictured scene/people and the video for actions, expressions and performance rhythm.

Missing files and open checks

• The source does not declare requiredReferenceCount; the number of listed slots is not a provider limit.

2 listed slots · 2 linked · 0 unlinked

Why it is included

Included to prevent treating every input video as a file to edit in place. This source describes transferring performance into an image-led scene; the audit does not judge how well it works.

Next preparation step

Use your own still and movement reference only if you can use both. State which gestures and which scene traits should be followed, then inspect the chosen provider’s supported input roles.

Input roles

The source instructions explicitly separate Video 1 camera movement, Image 2 character identity and Audio 3 vocal performance. All three slots have URLs.

Missing files and open checks

• The source does not declare requiredReferenceCount; the number of listed slots is not a provider limit.

3 listed slots · 3 linked · 0 unlinked

Why it is included

Included as a rare complete image/video/audio link set. This is a source example, not a verified voice-cloning capability or consent record.

Next preparation step

Prepare your own or licensed portrait, camera reference and vocal track. Confirm the provider supports this combination and obtain the permissions needed for each role; do not merge all three into one vague style input.

Input roles

Exactly one first_frame slot is listed and linked. originalMode is image-to-video; source.reportedMode is FL2VA. No last_frame slot is present in this record.

Missing files and open checks

1 listed slots · 1 linked · 0 unlinked

Why it is included

Included to show why a reported model-family label must not manufacture an absent end image. The official request source uses a mutable main URL, while this local metadata snapshot is hash-pinned.

Next preparation step

Use a single authorized starting image for a provider-supported image-to-video task. Keep rack focus and steam as separate observable requests; do not add a supposed official final frame.

Input roles

Video 1 and Audio 1 are linked, while Audio 2 has no URL. The source distinguishes source footage, reused background audio and a separate voice-timbre reference.

Missing files and open checks

• Audio 2 has no declared file URL.

3 listed slots · 2 linked · 1 unlinked

Why it is included

Included because two available files do not make a three-role bundle complete. Background music cannot silently stand in for the missing spoken-voice reference.

Next preparation step

Exclude direct replay until the missing reference and rights are resolved. Alternatively record a clearly consented original voice and build a new disclosed exercise with supported provider inputs.

Input roles

The reference list names Image 1 but gives no URL. The prompt calls it a starting frame; raw originalMode remains unspecified and reportedMode remains Unknown.

Missing files and open checks

• Image 1 has no declared file URL.
• Original generation mode is unspecified; navigation mode is not evidence.

1 listed slots · 0 linked · 1 unlinked

Why it is included

Included as an unknown-mode case with a clearly missing declared image. A descriptive prompt or 20-second title does not establish an executable mode or supported duration.

Next preparation step

Keep it as reading material until the creator’s mode and input file can be verified. For a new exercise, choose a duration and input type actually supported by your provider; do not inherit the title as a limit.

Input roles

Image 1, Image 2 and Audio 1 are named but all three URL fields are absent. The prompt describes character, environment and an editorial-timing music track.

Missing files and open checks

• Image 1 has no declared file URL.
• Image 2 has no declared file URL.
• Audio 1 has no declared file URL.
• Original generation mode is unspecified; navigation mode is not evidence.

3 listed slots · 0 linked · 3 unlinked

Why it is included

Included because a detailed multimodal plan can still be non-executable. The stored sourceId and source post URL use different numeric IDs; both are preserved for a reviewer to trace.

Next preparation step

Exclude direct replay, verify the original post and obtain an authorized three-input set first. Rebuild exact character names and timing labels as editorial graphics rather than promising model lettering.

Input roles

The reference array is empty and requiredReferenceCount is 0, but the prompt explicitly contains @[ref img] as the visual authority. No file URL is supplied for that mention.

Missing files and open checks

• The prompt names @[ref img], but no structured input slot or URL resolves it.
• Original generation mode is unspecified; navigation mode is not evidence.

0 listed slots · 0 linked · 0 unlinked

Why it is included

Included as a prompt/metadata discrepancy. Empty structured metadata does not establish a text-only source, particularly when the prompt names a reference.

Next preparation step

Keep mode unknown and flag the unresolved prompt reference. Request or create an authorized dossier image, then treat typography as editable layers; that creates a new exercise, not a verified reconstruction.

Input roles

Two reference_image slots, Image 0 and Image 1, are named but unlinked. The prompt calls them first and last states. raw originalMode is reference-to-video while reportedMode is FL2VA.

Missing files and open checks

• Image 0 has no declared file URL.
• Image 1 has no declared file URL.

2 listed slots · 0 linked · 2 unlinked

Why it is included

Included to preserve a label disagreement without silently relabeling the record. A two-state description cannot supply the missing files or verify a current end-frame feature.

Next preparation step

Exclude direct replay. Obtain authorized start/end states, document how they correspond and verify an actual end-frame interface before designing a new transition task.

Input roles

One reference_image slot is named without a URL; the prompt treats Image1 as the visual authority for the page. originalMode is reference-to-video and reportedMode is FL2VA.

Missing files and open checks

• Image 1 has no declared file URL.

1 listed slots · 0 linked · 1 unlinked

Why it is included

Included to show that a single named reference and a model-family label do not prove an endpoint pair. No first_frame or last_frame slot is declared here.

Next preparation step

Use only as a planning reference until the page image and usage rights are available. For your own illustration, preserve readable type in an editable overlay and keep the raw source mode unchanged in notes.