Every model. Every tool. Every tradeoff.

Strengths

• +Up to 30s in a single generation — multi-shot sequences without stitching or per-cut identity drift
• +Generates synchronised audio in the same latent space as the picture
• +Accepts image, video AND audio references together
• +Seven aspect ratios including 21:9 and 4:3

Weaknesses

• −720p ceiling on fal as of 2026-08-09 — 1080p and 4k were removed from the endpoint; use Seedance 2.0 for 4K
• −~1.5x the 2.0 token rate ($0.0214 vs $0.014 per 1k) — a 30s 720p clip is ~$14
• −No fast tier, so there is no cheap draft option on 2.5
• −Unproven here; we have not run a side-by-side against 2.0 yet

Strengths

• +Best face lock under heavy motion in our tests
• +Strong cinematic VFX and lighting consistency
• +Reliable identity preservation across takes

Weaknesses

• −Mediocre on hands when subject gestures
• −Backgrounds can drift on clips longer than ~6s

Strengths

• +Native 2K — not an upscale of 1080p
• +Generates its own stereo audio alongside the picture
• +Up to 15s in one take, where Seedance stops at 12
• +Cheaper per second than Seedance Standard despite the higher resolution

Weaknesses

• −No seed parameter — runs cannot be pinned, so one-variable A/B testing is not available
• −Reference images past the first five cost $0.08 each
• −New here: we have no measured face-lock or drift numbers of our own yet

Strengths

• +Cheapest cloud render available here — good for burning through takes
• +Identical model and prompt behaviour as the 2K entry, so what you fix carries over
• +Still generates its own audio

Weaknesses

• −768p — a draft resolution, not a deliverable
• −No seed parameter, same as the 2K entry
• −No measured quality numbers of our own yet

Strengths

• +~20% cheaper than Standard with most quality intact
• +Faster turnaround — good for A/B drafting
• +Same face-lock backbone as Standard

Weaknesses

• −Visible quality dip on complex multi-subject scenes
• −Texture detail softer than Standard at 1080p

Strengths

• +Free forever — runs on your own GPU, no per-second cost
• +No cloud queue — starts in seconds, not hours
• +Privacy: footage never leaves your machine
• +Better motion + identity than Wan 2.1 in our tests

Weaknesses

• −Still behind Seedance on demanding face-heavy shots
• −Hands and crowd scenes weaker than top cloud models
• −Requires ComfyUI install and 16+ GB GPU memory

Strengths

• +Free forever — runs on your own GPU, no per-second cost
• +Inserts you into existing footage at near-cloud quality without renting Runway/Pika
• +Locks identity across all frames of the clip via PuLID face conditioning
• +Source video can be anything — public-domain clips, your own footage, royalty-free B-roll

Weaknesses

• −Requires Wan 2.2 base model (Q5_K_M quantized version — compressed so it fits a 4090)
• −Hands and crowd geometry still slightly behind Seedance / Veo on tight close-ups
• −Source video length matters — best results at 3-8 seconds; longer clips drift on subject identity
• −Requires a clean reference photo of the target face (high-res, neutral expression, well-lit)

Strengths

• +Free forever — runs on your own GPU, no per-second cost
• +Only LOCAL option for first/last frame interpolation — Wan 2.1, Hunyuan, LTX do not have this capability
• +Locks both ends of the clip — model can't drift away from your end keyframe
• +Privacy: both source images and the output never leave your machine

Weaknesses

• −Requires Wan 2.2 base model — full fp16 doesn't fit most consumer GPUs; we ship the Q5_K_M quantized build (compressed version of the model that runs on smaller GPUs at slightly lower quality)
• −For single-frame I2V (one starting image only), Wan 2.1 wins on quality — swap back when you don't need the end-frame lock
• −Both source images MUST match framing/composition exactly — same camera, same crop, same subject position — or the morph looks broken
• −Caps at 5s output; longer morphs need to be chained

Strengths

• +Free forever — runs on your own GPU, no per-second cost
• +No cloud queue — starts in seconds, not hours
• +Privacy: footage never leaves your machine
• +Mature: well-tested workflow, lots of community fixes

Weaknesses

• −~70-80% quality vs Seedance on demanding shots
• −Hands and complex motion are noticeably weaker
• −Requires ComfyUI install and 16+ GB GPU memory

Strengths

• +Best-in-class motion realism among free local models
• +Body physics and limb movement noticeably better than Wan
• +Strong on action, walking, gestures
• +Different lineage from Wan — survives a broken WanVideoWrapper install

Weaknesses

• −Slower than Wan — heavier model
• −Identity lock weaker than Wan on tight close-ups
• −Larger download (~13 GB)

Strengths

• +Free forever — runs on your own GPU, no per-second cost
• +Only local model in this catalog with native synced audio + video in a single pass
• +Widest VRAM range in the catalog — fp8 runs at 12 GB, GGUF builds go lower
• +Strong fine-tuning ecosystem: IC-LoRA adapters, first/last-frame and camera-control LoRAs

Weaknesses

• −Native audio is impressive but not broadcast-clean — expect to replace VO on anything scripted
• −Face lock still behind Seedance on tight close-ups
• −The quantization spread is a trap — grab the wrong build and quality drops hard with no warning

Strengths

• +Distinctive cinematic look — strong on establishing shots
• +Good camera-move physics (push-ins, dolly drifts)
• +Lighting feels more 'film' than other cloud models
• +Roughly a third the per-second cost of Seedance Standard

Weaknesses

• −Identity preservation drifts on tight close-ups
• −Image-to-video only on fal — no multi-reference input like Seedance

Strengths

• +Best-in-class motion realism for action / sports
• +Handles fast camera moves without warping
• +Crowd / multi-subject scenes hold up

Weaknesses

• −Face lock weaker than Seedance on close-ups
• −Inconsistent lighting frame-to-frame on long clips

Strengths

• +Reference-driven animation — strong from a single image
• +Storyboard-friendly: respects pose / motion guides
• +Good for explainer / illustrative work

Weaknesses

• −Less photoreal than Seedance on talking heads
• −Pricing and slug not yet verified — treat as experimental

Strengths

• +Best-in-class text-to-video coherence
• +Cinematic look with very good world physics
• +Native synced audio on supported endpoints

Weaknesses

• −Reference-image fidelity weaker than Seedance
• −Identity-lock not as tight on returning characters

Strengths

• +Cheapest premium cloud model in the catalog — $0.10/sec
• +Google's own recommended default over VEO 3.1
• +Native synced audio
• +Built for conversational video editing, not just one-shot generation

Weaknesses

• −Public preview — quotas, pricing, and region availability can change without notice
• −Caps at 10-second generations for now
• −Google-direct only: does not ride our existing fal.ai plumbing
• −Preview model IDs get retired; expect to re-pin the slug