Wan rents the film. LTX-2.5 keeps the cut

Wan 3.0 is generally available, and it is still rent. Up to thirty seconds in one pass. Omni-reference that now includes documents, slides, and web pages, not just pretty stills. Edit carry-forward so a change does not throw the whole concept away. Per-second API pricing you can put on a spreadsheet. No weights. Last December we said Wan 2.6 was the clip everyone forwarded and 2.2 was what you hosted. This month did not retire that sentence. It made the rented half look like a brief you already had.

That is the honest pitch. Drop a deck on the model and ask for a launch film. Sometimes you will get a first assembly that would have taken a week of translation. Sometimes you will get a pretty misunderstanding of slide eleven. Either way, you are paying by the second, and you are not taking the engine home.

A model that can read the deck is a collaborator. A model that will not send you the checkpoint is still a vendor.

The rented film that finally takes the brief

Thirty seconds matters if the face holds. It does not matter if shot two hired a different actor. Alibaba is claiming identity, voice, and lip sync as first-class work. Audio texture and on-screen type are still the soft spots. Believe the first claim after you run your own product. Treat the second as a known repair.

Omni-reference is the part a marketer will feel. A campaign rarely starts as a prompt. It starts as a deck, a grid, a product page, a voice note. A model that can take those files reduces the intern-work of restating the brief in adjectives. It does not reduce the need for a person who can say the offer is wrong.

Pricing is refreshingly dull. About five cents a second at 480p, ten at 720p, twenty at 1080p. A thirty second finish at 1080p is about six dollars before anyone talks you into another take. Draft cheap. Finish on purpose. That is a real workflow, and it is still a rate card.

Can you automate it

Yes, if the call returns a file and a receipt. Deck in, clip out, check the face, check the type, refuse the take that invented a feature. An agent can queue that. A person still has to watch the ones that almost work.

Do not automate the upload of every slide you have ever made. Automate the three tests: a twenty second narrative, a reference that must hold, an edit that changes one line. If those three fail, you do not have a production model. You have a mood board with an invoice.

The local cut that learned multi-shot

LTX-2.5 is the open-weight answer sitting on the other bench. Native multi-shot. Stronger prompt adherence. A better decoder. A distilled path that can run faster than real time on serious iron. Same commercial ceiling as the rest of the LTX family: under ten million in revenue, you are fine. Over it, you buy a seat. We already wrote the operator's version in LTX-2.5 Pushes Open-Weights AI Video Toward Real Workflow Automation. The month-end version is the one you tell a producer. You can hold a scene across cuts on a box you already have.

That is the job LTX has been chasing since January. 2.3 made the desktop feel like a product. 2.5 makes the scene feel like a scene. If you spent the first half of the year on 2.3, this is the upgrade, not a new religion.

It is still not a mix engineer, and it is still not a legal department. Stems arrive. Faces hold more often. You still need ears, and you still need a person who can say the claim is wrong.

A character you can drive

Wan-Animate-2 is the open side of the Wan house this month, and it is a different product than 3.0. Apache 2.0. Fourteen billion parameters. A reference image, a driving video, a character that keeps the face. No audio. This is performance capture without the mocap stage, not a thirty second ad engine. Host 2.2 if you want the older talking stack. Call 3.0 if you want the rented film. Download Animate-2 if you want to put a kit character on a motion you already shot.

Do not mix the names in a slack thread and then wonder why the output has no voice. They are cousins. They are not a bundle you own.

Runway Aleph 2.0 is the hosted edit. We covered it when it landed: Aleph 2.0 Makes AI Video Editing More Workflow-Ready. Useful when the job is "change this, keep that" and you have accepted the invoice. Not a replacement for LTX. Not a replacement for a person with a timeline.

What a marketer can actually plug in

Wan 3.0 when the brief is a deck and the file can die on their servers. LTX-2.5 when the film has to live on yours. Animate-2 when you already have a character and a drive. Aleph when the edit is the product and Runway is already in the stack.

Four paths is too many for a Tuesday. Pick two. One you keep. One you rent. Write the rule on the wall. The rest of the year will try to sell you a third.

Wan 3.0 did not send you a checkpoint

Wan 3.0. Aleph. Seedance. Veo. Anything that will not send you a checkpoint. LTX-2.5 and Animate-2 are the exceptions, each with a licence you should read before a client asks. Wan 2.2 is still the hostable Wan flagship. 3.0 did not change that, no matter how good the deck-to-film reel is.

Pick two paths, write the rule

One local multi-shot path. LTX-2.5 if you can hold it. One rented film path only if you have a rate card and a reject list. One character driver only if you already have a character.

Do not rebuild the stack because a model learned to read PowerPoint. Reading the deck is a gift. Keeping the engine is a decision. We have been making that decision since December. This month just made both sides look like grown-up products.

GroundSlate is where the local half sits. The multi-shot. The character. The brief you still write yourself. The rented half is a line item. August made that line item harder to ignore, and easier to price.