<hr />13b Betaltx 13bFirst let me say, this is ... not fun to use.It uses 10+gb of vram to get 1024 output. Its faster than others at doing this but its very hard to control and so many options are hidden behind configs and manual edits.. I cant recommend this to anyone looking for simple image to video.That being said, it works and can do decent for some things but output seems random. One video is static image, next is a blurry mess of movement, you seem to need to change seed 3-5 times before you find something that is ok.<hr />If you are looking for image to video i currently use wan and framepack both work on low vram and offer better results than LTX currently.. Lets hope 1.0 does something to fix that...WAN 2.1 1.3b INP model (first and/or last frame)<a target="_blank" rel="ugc" href="https://civarchive.com/models/1300494/wan-21-video-generation-img2vid-with-fun14b-image-to-video">https://civarchive.com/models/1300494/wan-21-video-generation-img2vid-with-fun14b-image-to-video</a>Framepack (first and last frame)<a target="_blank" rel="ugc" href="https://civarchive.com/models/1492481/framepack-image-to-video">https://civarchive.com/models/1492481/framepack-image-to-video</a>V9.1-Added guided insert option to base-removed crystools as a dependent node set-added note with needed nodesother flows- no LLM searge version for those that dont want to use searge(understandable)- full basic (this should work with minimal nodes)Including- reactor face flow- combine flowenable previews of the animated videos in the samplers in comfy optionsunder VHS - Change display animated previews while sampling. This lets you cancel videos that go off the rails easier.** if you are trying back to front animations (last frame insert) you should connect the frame to insert to node to the upscale guided nodes frame_idx to upscale it properly.** new model tends to do quite well upscaling with 10 steps.<hr />V9 v0.95 - the bullshit release (hotfixed 2 small bugs same download)Extensions and fixes for V8.1 (phoenix release)My attempt to bring back SOME of what made v8 good.Model insert is back for the base render. We need latents now so its not perfect but mostly works like V8. I don't like stg, so perturbed is back.-Removed 2nd image-Changed inserts to new ways. (works but slightly worse than v8.1)-Changed upscale to new ways. (works and might be better than before v8.1)-model change option for upscale (0.91 base to 0.95 upscale works)-Not using STG... I hate STG.... really hate it.-Perturbed is back.-masked noise removal back, all noise options back-all prompt options back (delete any groups you don't want)-added optional options groups from new flow-added optional reactor face swapworking with 0.91 and 0.95<hr />Alternate WAY<hr />v1.01 hotfix - fixed upscale ref sizev1.02 hotfix - No changes, just removed F5 and deepextract nodes from main flow (kept it there if wanted in v1F5TTS and added the combine flow and reactor face flows.Original seed of the walking girl 364851990605292 and it was area resize.If having memory problems when using face restore, connect unload reactor models after the face restore, i forgot to add that.Bugs ** to be fixed soon (non breaking)<ul><li>The teacache needs to have its nodes reversed. Overrider needs to be after the cache. This makes it work correct, but can mess up videos if set too high.</li><li>Model for upscale needs to be connected from STG in upscale to the STG guider</li></ul>Sorry if you installed F5 and did not need to.I guess this is how we doing things nowV1 reborn<ul><li>added teacache</li><li>added crop nodes</li><li>added optional groups to end</li><li>added noise option for upscale</li><li>Tried to unify most nodes, full list of nodes/models needed in notes.</li><li>Few other qol changes.</li><li>added notes</li></ul>fixed image to vid links.<hr />0.95 betatester - i really miss the old nodes....Bug - Noise after resize should be connected to image to video nodeThis is an entirely new flow that started as a test flow for new sampling methods for ltx.No notes, very raw. Should be easy enough to follow.Single image, 2 insert ways, upscale test.THe main things to use are CFG and STG levelslower stg for more movement higher for less. and like perturbed rescale tells it when to stop.Noise does not control much at all with the new nodes and model.<hr />Bugs - groups for noise for second insert need to be renamed to use 2 instead of 1 to fix the bypasses to work properly.If looking for my reactor face swaps and combine workflows its still in v8. I did not add them atm to 8.1V8.1 - <a target="_blank" rel="ugc" href="https://huggingface.co/Lightricks/LTX-Video/tree/main">https://huggingface.co/Lightricks/LTX-Video/tree/main</a>90% sure this can make end frames usable to extend...The limit of 1024 frames tall seems lifted with this version.tested 305 frames, it did quite well. Not sure on limits.older models work fine in the flow. 0.91 still can be usefulYou can try inserting only last frame, it can backwards render. Just not quite as well.This workflow works without the upscale if you want. its can do 720p or more in base.Upscale can go up to 1080p from that (my vram uses 80% of my 12gb for 1080p upscale)<ul><li>Updated to 0.95 model and nodes</li><li>added latentsync into flow</li><li>put remove frames at end of flow</li><li>Added new LTX noise node option</li><li>2 image interpolation working quite well.</li></ul><ul><li>Removed all old latent guide nodes, replaced with new ones.</li></ul><hr />For those that are trying to get V8 working after updating.Feta enhance needs to be removed or it will error with (shape)?modify LTX model may not be needed.Imagetovid conditioning insert is borked with updateInserting via model MUST insert latents now. This seems to insert properly to the first/last you choose unlike new conditioning latent guilds.Upscale needs shift node removed(i dont know where it went or a replacement)<hr />V8 - beta - i hope this is better with more options, but i cant test everything. If you find a bug or something you want back from v7 let me know in comments.- added multiple new noise options- changed resize of ref images options- added masked noise removal- removed 2 perturbed layers- optimized prompting- optimized after video options- added insert via conditioning (teacache enabled) teacache and LTX dont mix well- added face fix group to after options.including reactor face fix/change flow if needed.Having issues, try NO prompt.See what the image does without a prompt, noise or not can force movement, the video tends to follow whatever it sees in the image.<hr />Info here.<a target="_blank" rel="ugc" href="https://civarchive.com/articles/10440">https://civarchive.com/articles/10440</a><a target="_blank" rel="ugc" href="https://civarchive.com/articles/11073">https://civarchive.com/articles/11073</a>The blur noise group node in the flow for both images can be changed.<img src="https://image.civitai.com/xG1nkqKTMzGDvpLrqFT7WA/d80cc80e-4a84-4ed2-986b-2becebd60699/width=525/d80cc80e-4a84-4ed2-986b-2becebd60699.jpeg" />Sigma x is left right blursigma y is up down blur.x seemed to work similar to noise if you want to try using only it. 0.25-.75V7Change to how things render.- New option to use inserted latents or emptywhen using empty (new default) This unlocks- Clearer Base videos- Clearer upscales- easier extends (still not perfect, but the clearer base/upscales helps) color match is less needed unless above 1024- Slightly larger upscales above 1152 i had slight blur issues. This may be able to be solved with upscale settings but i have not found anything that can clear up past this point.- Smaller videos work well. 320x256 animates fine, and upscales 2x without problems and can flirt with doing 3x.- Cleaned up the last of the node line spam.All options still work with default empty latents, noise, first/last except add new frames. You need to insert to do that. And it works the same as before.First last options seem to works better without latent inserts.Same image in first/last generates loops, ususally needs noise to force it to move reliably. Backgrounds sometimes dont animate properly with thisIf you want an idle animations and they wont stand still for you. first/last as same image works well if you add enough noise to the first image to animate between the 2 frames.- added image selecter to image load to allow animated gif loading.- this can be used as text to video without the inserts enabled.<hr />

LTX Image to video

4.3 - small update- changed vae decodes/encodes to tiled i dont notice any diff between this and normal besides it does not lock my pc when decoding large videos. You can change this if you prefer normal ones. (note - temporal size is max. It causes a skip in the video if not)- Researched Perturbed attention - new base settings and explanation- added color match node at end of base and upscale to attempt color match the video properly back to the original image.- added main ref tagging option for extends.- Fixed vae swap i did not know did not work. Added VAE decoder noise node.

ltxImageToVideo_v43.zip

v4.3

13b Beta

v9.1

v9 Phoenix

Reborn V1.02

beta

v8.1 (0.95)

v6.1

SettingsTester

v5.1

ExtendsBeta

v4.2

v4.1

v1.1

v1.0

Files with SHA256: bbb651028733568808ea66aa0a89055487407cd9fc532cc3b9493b3176cc53dc