# Kling Lab: First Clip from a Finished Frame A strong frame has already solved much of the task: the character is recognisable, the clothing is correct, and the lighting works. When you start moving that frame, a face may change, an umbrella may acquire a twin, and the camera may decide to direct its own film. That is why the first test is deliberately small. You will use **one finished frame**, set **one visible action**, and assess **one short clip** against a criterion written in advance. The goal is not to guess every capability of the current interface. It is to create an honest, comparable test. ```learner-callout {"stable_id":"c181-key-small-test","role":"key-point","title":"One frame, one action, one criterion","body":"The start frame controls appearance. A short movement description controls the change over time. The actual clip is the only evidence of what happened."} ``` By the end of the lab, you will have an image-based route confirmed at the time of work, or the honest status “prepared to run,” plus a start frame, two comparable versions, and an `accept / retry` decision. ## 1. What Kling Is and Where I2V Fits In this lab, **I2V** means a route that accepts an image as starting material and creates a short moving segment from it. Tab names, available settings, and the path through the interface can change. Follow the input rather than an attractive label: the selected route must let you upload a frame and receive video. | Item | Plain meaning | | --- | --- | | Kling | the named service in which the video-generation route is checked | | interface | the current page or app with fields and controls | | I2V | a route where an image is the starting material | | prompt | text about movement, camera, preservation, and a visible ending | ```learner-callout {"stable_id":"c181-source-runtime-boundary","role":"source","title":"Verify the current route","body":"Local sources support the initial I2V approach and prompt pattern, but do not prove today’s labels, settings, access, pricing, or result. Record what is available in the interface on the day of the test."} ``` For a first run, you need access to the current interface, one image in a format it supports, one simple action, and time to watch the full result. ## 2. Assign Responsibilities In I2V, the jobs are divided. The start frame already carries the character, clothing, objects, location, composition, and initial light. Text is not for repeating every pixel; it is for movement, camera, preservation of what matters, and an end state. ```text frame controls appearance -> prompt controls movement -> clip shows what actually happened ``` ```learner-callout {"stable_id":"c181-technique-frame-prompt-division","role":"technique","title":"Do not describe the frame again","body":"If the yellow raincoat is already visible, do not say it five times. Use the text for one action, camera behavior, important preserved features, and the clip ending."} ``` This division does not guarantee preservation. It makes a failure diagnosable: you can see whether an object changed, the camera drifted, or the action did not finish. ## 3. First Run: from Input to Result Use this practice scene: a woman in a yellow raincoat stands under a shelter and holds a closed transparent umbrella. She opens it and takes one step into the rain. 1. Open the current official Kling interface and its video-creation area. 2. Find a route that accepts a start image and record its actual name. 3. Upload the accepted frame of the woman and umbrella. 4. Select an available short duration and an aspect ratio suitable for the source; record the values rather than assuming them in advance. 5. Use the first test text. 6. Run the generation if access and the route are confirmed. 7. Open the complete clip, not only its attractive first frame. 8. Check the face, raincoat, umbrella, camera, and final state. 9. Save the file under a clear name and accept or reject it. **First-run prompt:** ```prompt The woman opens the transparent umbrella and takes one deliberate step from under the shelter into the rain. The camera makes a slow, steady lateral move that keeps her face, yellow raincoat, and umbrella readable. Preserve the same person, clothing, shelter, street geometry, rainy lighting, and umbrella design from the attached start frame. End with the umbrella fully open above her as she stands in the rain. No extra people, no duplicated umbrella, no sudden camera shake. ``` ```learner-callout {"stable_id":"c181-observation-watch-whole-clip","role":"observation","title":"Check movement, not the cover","body":"A successful test is the same person opening the same umbrella, taking one step, and ending under the open umbrella. If the face changes, the umbrella duplicates, or the action stops early, the test fails even when the rain looks beautiful."} ``` ## 4. Build Your Own I2V Prompt A short test needs five parts: ```text 1. subject movement 2. camera behavior 3. lighting or environmental response 4. what must remain consistent 5. visible ending and exclusions ``` Write what can be seen. “She feels free” gives no testable movement. “She opens the umbrella, takes one step, and stops” does. Before the text, complete this short card. It is a form for your own test, not a promise of an interface capability. ```template Start frame: What is already correct: One main action: Camera behavior: What must remain consistent: How the clip ends: Observable fact that makes me accept the result: ``` If one short segment asks the character to stand, open a door, turn, speak a line, and run into the rain, you are testing five jobs at once. Split them into several clips. ## 5. Second Test: Change One Thing To understand the cause of a result, keep the frame, action, duration, and ending the same in the second version, but stop the camera. ```prompt The woman opens the transparent umbrella and takes one deliberate step from under the shelter into the rain. The camera remains completely static and keeps her face, yellow raincoat, and umbrella readable. Preserve the same person, clothing, shelter, street geometry, rainy lighting, and umbrella design from the attached start frame. End with the umbrella fully open above her as she stands in the rain. No extra people, no duplicated umbrella, no sudden reframing. ``` ```learner-callout {"stable_id":"c181-inference-one-variable","role":"inference","title":"A comparison must answer one question","body":"The question here is: does a static camera preserve readable action better than a lateral move? If you also change the frame, speed, light, action, and camera, you get two different videos but no explanation of the cause."} ``` Write the difference between versions in one line: `lateral move → static camera`. Keep the other fields unchanged as far as the available route allows. ## 6. Accept or Reject the Clip Check the task before the beauty. ```template Face and raincoat stayed consistent: yes / no Umbrella remained the same object: yes / no Action reads from start to finish: yes / no Camera behaves as requested: yes / no Final state was reached: yes / no Decision: accept / retry One cause for the next correction: ``` Beautiful rain cannot rescue a clip with a duplicated umbrella. Strong camera movement is not a win when the main action leaves the frame. If the result does not pass, change one cause: shorten an overloaded action, name a preserved object more clearly, keep an interfering camera static, simplify the ending, or stop and find another confirmed route if the selected one does not accept the needed input. ## 7. When Simple I2V Is Not Enough Some tasks need another input. An exact final frame requires checking a separate start-and-end-frame route; transfer of existing movement requires confirming availability and the way motion control accepts input; several scenes require splitting into simple clips or a confirmed dedicated route. ```learner-callout {"stable_id":"c181-caution-no-interface-inference","role":"caution","title":"Do not infer a function from an old example","body":"An old screenshot and someone else’s successful prompt do not prove current feature availability. Before running, confirm the needed input in your interface; after running, only a reviewed file against the written criterion becomes evidence."} ``` Do not turn hope in text into a substitute for the appropriate input. If the needed route is absent, the honest lab result is a prepared test and recorded stop, not an invented clip. ## 8. Practice and Lab Completion 1. Choose your own start frame. 2. Name one main action and a visible ending. 3. Find an image-based route in the current interface and record its name and check date. 4. Prepare the base test. 5. Prepare a second version that changes only the camera or the speed of one action. 6. If running is available, save both files and compare them against one criterion. ```learner-callout {"stable_id":"c181-practice-honest-status","role":"practice","title":"Two correct statuses","body":"If files exist, watch them fully and accept or reject them with the card. If access or execution is unavailable, record: “route and prompt prepared; generation not run.” Preparation for a test is not presented as completed generation."} ``` ### Check before Leaving ```learner-callout {"stable_id":"c181-checkpoint-lab-ready","role":"checkpoint","title":"The lab is completed honestly","body":"There is a start frame, name and date of the checked route, two versions with one stated difference, and a decision criterion. When files exist, they were watched completely; without files, the status remains “prepared to run.”"} ``` Lesson result: do not ask a good frame to invent itself again. Let the image control appearance, the text control one movement, and the actual clip provide evidence of the result.